Skip to content

Moonshot AI · Chat

Kimi K3

Moonshot AI 🇨🇳ChatKimi familyOpen weights. One of its sources says the weights are publicly available.The maker currently plans to deprecate it within the next 180 days (14 Dec 2026). Plans can change.

Newer version: Kimi K3.1. Our catalogue has no release date for Kimi K3.1. We compared the version names in our catalogue.

From Empiriolabs's listing:
Kimi K3 is Moonshot's flagship reasoning model with a 1M token context, always-on thinking, native web search, and text, image, and video inputs.
llminfo IQ
139.2
Maker's price
$6.00 blended / 1M
Max context
1.1M
Listings
234

Evidence and sources

International API storefront is Singapore-based; principal operating entity 北京月之暗面科技有限公司 (Beijing).

🤗 moonshotai/Kimi-K3 Model architecture, as its sources state it llminfo IQ 139.2, ranked among chat models. From 417 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 123.8, 95% interval 117–131. From 25 benchmark families across 12 sources. Plenty of data behind it. Uses the model's best published results. It has its own scale, separate from llminfo IQ.How Coding IQ is computedGood value (60.2). Cheaper per task than its benchmark results predict. From 77 benchmarks with measured task costs. Plenty of data behind it.How Value for money worksYap score: tokens used per coding task vs. models with the same resultsTypical (55 ±9). Generates about as many tokens as models with similar coding results. Measured at max effort. Little data behind it, so treat it as a rough estimate. By effort: default 58 · none 57 · max 55.Median 195,427 output tokens per task at max effort.How the Yap score worksReleased 16 Jul 2026Maker HQ: CN (source)
All rankings: #21 on OpenRouter this month #30 on OpenRouter this week #44 on OpenRouter today #63 on NanoGPT recentlySources: Arcee AI data, EmpirioLabs data, LiteLLM data, Models.dev data, Morpheus Inference Marketplace data, llminfo researchPrimary data from Arcee AI from EmpirioLabs: primary mode, description from LiteLLM: deprecation date from Models.dev: weights availability from Morpheus Inference Marketplace: architecture from llminfo research: release date
Tags from its sources:instructtoolsreasoningvisionopen-sourcemoelong-contextcodingfunction-calling

Advantages and trade-offs

Compared with 10 chat models, picked mainly by popularity and IQ

Advantages

Open weights available

The weights are public. 2 of the 9 peers that say either way have open weights.

Trade-offs

Deprecation scheduled

The maker plans to deprecate it on 14 Dec 2026, within the next 180 days.

We only make a claim when there's enough data on models of the same type. Updated 22 Nov, 04:34 UTC.Prices come from the model's maker where they sell it, otherwise their official partner, otherwise the market median. This comparison uses 10 maker's prices.

Input & output

as stated by the model's providers
Text
Images
Video
Files
Text

Capabilities

Supported operations and provider-stated features

VisionReasoningFunction callingStructured outputPrompt cachingWeb searchBatch API

Listings

 Moonshot AIkimi-k3Sources: GenAI Prices data, LiteLLM data, Models.dev data1M$3.00cache read $0.30$15.00cache write 5m $3.00cache write 1h $6.00——
 Aihubmixcoding-kimi-k3-free · coding, freeSource: AIHubMix data docs1M$0.00$0.00——
 Qiniumoonshotai/kimi-k3-free · freeSources: Qiniu AI data, Qiniu AI Model Plaza Pricing1M——
 OpenRouterInference by Wafermoonshotai/kimi-k3 · fast · mxfp4Source: OpenRouter data docs1M$0.153cache read $0.152$16.0095640.5ms
 Hugging FaceInference by Relacemoonshotai/Kimi-K3Source: Hugging Face Inference Providers data1M$0.1824$11.006——
 NanoGPTInference by RelaceRegions: UNKmoonshotai/kimi-k3 · fp4Source: NanoGPTNot published by the provider$0.1915cache read $0.189$11.55690.61,851ms
 Flowdotaccounts/fireworks/models/kimi-k3Source: FlowDot data1M$0.50$0.50——
 WaferKimi-K3 · mxfp4Sources: Models.dev data, Wafer data1M$0.60cache read $0.20$15.00——
 Uplinkswitchkimi-k3Source: Models.dev data1M$0.90cache read $0.12$4.50——
 Nous Portalmoonshotai/kimi-k3Source: Nous Portal Inference Gateway data1M$0.99cache read $0.20$8.99——

16 listings without a price are hidden.

Included in plans

Umans Code ProumansSubscription$20/moslows down past the allowancePlan page
Umans Code MaxumansSubscription$50/moslows down past the allowancePlan page
Copilot EnterprisegithubSubscription$39/moextra use billed past the allowancePlan page
Copilot ProgithubSubscription$10/moextra use billed past the allowancePlan page
Copilot Pro+githubSubscription$39/moextra use billed past the allowancePlan page
Copilot BusinessgithubSubscription$19/moextra use billed past the allowancePlan page
Copilot MaxgithubSubscription$100/moextra use billed past the allowancePlan page

Benchmarks

Named effort levels run from lowest to highest. Select a result to see its source details.

Agentic

Reported by Z.ai, DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Kimi Code · official leaderboard snapshot 2026-07-23
Arena AgentSource: LMArena data
BFCL-V4Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · official BFCL V4 harness commit 6ea5797 · 5,217 cases · temperature 0.001 · weighted overall score · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
BrowseComp4 results
Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · in-house search/visit/python harness · 500 steps · discard-all context strategy
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
context compaction at 300K · 1M context · max effort · temp 1.0 · top_p 1.0
no context management · full 1M context · max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
DeepResearch Bench IIAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · Claude Code · identical search tools · four-hour task limit · GPT-5.5 judge · three independent scoring passes · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
GAIA4 results
Source: GAIA Leaderboard data · Sep 2026
Difficulty level
Difficulty not reported
GDPval · Atria paper EloAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 220 tasks · AgentCompass / OpenHands · four-hour limit · 250 steps · GLM-5.3 rubric judge · pairwise Elo · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
GDPval-AA v22 results
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Artificial Analysis / Z.ai转载
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
snapshot 2026-07-23
Harvey LAB-AAcriterion pass rate · snapshot 2026-07-23 · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
JobBench3 results
Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · official OpenCode harness · main split / 65 tasks · 7,200-second limit · two attempts · Grok-4.3 judge · mean task score
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
MCP Atlas2 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
500-task public subset · 100-turn limit · Gemini 3.1 Pro judge
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
MCPMark VerifiedSource: MCPMark Verified leaderboard docs
OSWorld 2.0max effort · temp 1.0 · top_p 1.0 · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official Harbor · H100 · max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SWE-Bench ProAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · OpenHands v1.23.0 · 4.5-hour limit · 250 interaction steps · network disabled and Git history removed · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: OpenRouter benchmark exports
15 Jul 2026 version · 90-day results
Across 23 providers lowest 71.3% median 74.1% highest 76.2%OpenRouter's automatic routing 74.5%
Show results from 23 providers and OpenRouter's automatic routingHide results from 23 providers and OpenRouter's automatic routing
  • BaseTen
  • AkashML
  • Nebius
  • ParetoAlpha
  • InferenceNet
  • Wafer
  • Decart
  • Io Net
  • Phala
  • OpenRouter's automatic routing
  • Alibaba
  • Relace
  • Modal
  • DeepInfra
  • Morph
  • Makora
  • Parasail
  • Moonshot AI
  • Together
  • Sail Research
  • Fireworks
  • DigitalOcean
  • Amazon Bedrock
  • KT
Terminal-Bench 2.14 related benchmarks · 10 results
Terminal-Bench 2.1
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Setting not stated
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
best score across harnesses · max effort
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Terminal-Bench 3.0
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Setting not stated
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek Harness / max effort
DeepSeek V4.1 launch chart · settings unspecified
Terminal-Bench 4.0
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek Harness / max effort
DeepSeek V4.1 launch chart · settings unspecified

Human preference

Arena ELO52 related benchmarks · 52 results
Source: LMArena data · Nov 2026
BenchmarkMax
Arena ELO
Chinese
Creative Writing
English
Show 48 more benchmarksHide 48 benchmarks
BenchmarkMax
Exclude Ties
Expert
French
German
Hard Prompts
Hard Prompts English
Industry Business and Management and Financial Operations
Industry Entertainment and Sports and Media
Industry Life and Physical and Social Science
Industry Mathematical
Industry Medicine and Healthcare
Industry Software and It Services
Industry Writing and Literature and Language
Japanese
Korean
Longer Query
Multi Turn
Non English
Polish
Russian
Spanish
Style-Controlled
Text Factuality
Text Factuality - Chinese
Text Factuality - Creative Writing
Text Factuality - English
Text Factuality - Exclude Ties
Text Factuality - Expert
Text Factuality - French
Text Factuality - German
Text Factuality - Hard Prompts
Text Factuality - Hard Prompts English
Text Factuality - Industry Business and Management and Financial Operations
Text Factuality - Industry Entertainment and Sports and Media
Text Factuality - Industry Life and Physical and Social Science
Text Factuality - Industry Mathematical
Text Factuality - Industry Medicine and Healthcare
Text Factuality - Industry Software and It Services
Text Factuality - Industry Writing and Literature and Language
Text Factuality - Japanese
Text Factuality - Korean
Text Factuality - Longer Query
Text Factuality - Multi Turn
Text Factuality - Non English
Text Factuality - Russian
Text Factuality - Spanish
Design Arena EloSource: Design Arena Leaderboards data · Nov 2026

Coding

Arena ELOSource: LMArena data · Nov 2026
CyberGym3 results
Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · Claude Code v2.1.119 · high thinking effort · four-hour-ten-minute limit · 2 CPUs / 4 GB
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Claude Code / 无Web工具
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
ExploitBenchReported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
FrontierSWE2 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Kimi Code · Dominance · snapshot 2026-07-16
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
FrontierSWE v2Proximal Proximus harness · mean@5 · max effort · Reported by SpaceXAI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Kimi Code Bench 2.0max effort · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SkillsBenchLing-3.1-flash official chart · SkillsBench · version, task count, harness, reasoning setting and run count not stated · Reported by Ant inclusionAI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
SWE-Marathon2 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
H20-calibrated pre-v1.1 task snapshot · max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Reasoning

Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Setting not stated
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
600-task public subset · max effort · temp 1.0 · top_p 1.0
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
AutomationBench v1.0.6Atria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · v1.0.6 with null-type fix from PR #13 · candidate harness and task count unstated · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
BabyVision3 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified · with tools
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Python tool · three-run average · max effort
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
ChartographyDeepSeek V4.1 launch chart · settings unspecified · with tools · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DECK Bench2 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
N/A harness · max effort · internal
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
DeepSearchQA2 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified · full set
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
full set · no tools · max effort · temp 1.0 · top_p .95
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
full set · tools · max effort · temp 1.0
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
MMVUthree-run average · max effort · temp 1.0 · top_p .95 · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
OfficeQA Pro2 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
max effort · temp 1.0 · top_p 1.0
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Verified · max effort · temp 1.0 · top_p 1.0
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
WorldVQA2 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
ForceAnswer prompting · three-run average · max effort · temp 1.0 · top_p .95
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
ZEROBenchSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Multilingual

EKI EstonianSource: Keelemudelite mõõdupuu data docs
NepNewsBench v2 draft40 related benchmarks · 40 results
entity F1 (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
core fact recall (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
core fact recall (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
core fact recall (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
coverage precision proxy (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
Show 34 more resultsHide 34 results
coverage precision proxy (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
coverage precision proxy (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
effective score (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
effective score (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
effective score (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
English prose quality (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
English prose quality (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
English prose quality (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity F1 (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity F1 (mixed English and Nepali source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity F1 (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity precision (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity precision (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity precision (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity recall (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity recall (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
entity recall (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
Nepali prose quality (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
Nepali prose quality (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
Nepali prose quality (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
supporting fact recall (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
supporting fact recall (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
supporting fact recall (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
valid-answer quality (all questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
valid-answer quality (English-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off
valid-answer quality (Nepali-source questions)
Source: NepNewsBench v2 (draft candidate release) data · Sep 2026
Moonshot AI via openrouter, reasoning off

General

Science

GPQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
GPQA Diamond32 results
Source: Epoch AI Benchmarking Hub data
Reasoning effort
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
Pass@1 / max effort
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
max effort · temp 1.0 · top_p .95
Source: OpenRouter benchmark exports
15 Jul 2026 version · 90-day results
Across 25 providers lowest 89.4% median 92.3% highest 93.1%OpenRouter's automatic routing 92.2%
Show results from 25 providers and OpenRouter's automatic routingHide results from 25 providers and OpenRouter's automatic routing
  • Io Net
  • Parasail
  • Nebius
  • Morph
  • Fireworks
  • Decart
  • Relace
  • River
  • Wafer
  • Makora
  • InferenceNet
  • Sail Research
  • Modal
  • ParetoAlpha
  • OpenRouter's automatic routing
  • Phala
  • BaseTen
  • KT
  • Moonshot AI
  • Amazon Bedrock
  • Chutes
  • Alibaba
  • AkashML
  • Together
  • DeepInfra
  • DigitalOcean

Long context

Math

MathArena ApexDeepSeek V4.1 launch chart · settings unspecified · Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
MathVision3 related benchmarks · 3 results
MathVision
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
With CI
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Python tool · three-run average · max effort
Without CI
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
no tools · three-run average · max effort · temp 1.0 · top_p .95

Vision

CharXiv-R3 related benchmarks · 3 results
CharXiv-R
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
CharXiv (RQ) · With CI
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Python tool · three-run average · max effort
CharXiv (RQ) · Without CI
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
no tools · three-run average · max effort · temp 1.0 · top_p .95
MMMU-Pro3 results
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
no tools · three-run average · max effort · temp 1.0 · top_p .95
Python tool · three-run average · max effort
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Video-MMEwith subtitles · three-run average · max effort · temp 1.0 · top_p .95 · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.

Factuality

Instruction following

Knowledge

OCR

Safety

Speech

Other

Artificial Analysis10 related benchmarks · 11 results
Briefcase
Source: LLM Stats data
Setting not stated
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
snapshot 2026-07-23
Show 4 more resultsHide 4 results
BlenderWorld15 related benchmarks · 15 results
Art & Animation
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 30 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Architecture & Interiors
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 72 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Characters & Animals
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 25 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Fully satisfied rubric
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 223 tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Image/video-input overall score
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 145 image or video reference tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Nature & Landscapes
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 21 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Show 9 more resultsHide 9 results
Overall score
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 223 tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Procedural & Game Assets
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 10 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Products & Manufacturing
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 28 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Rendering & Simulation
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 17 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Science & Visualization
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 9 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Text-only overall score
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 78 text-only tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Vehicles & Machinery
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 11 category tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Visual-quality dimension score
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 223 tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
Weighted rubric score
Reported by BlenderWorld authors / Alibaba Token Hub · via BenchBoard docs
benchmark-owner paper run · 223 tasks · Blender 4.2.10 / Blender MCP paper setup · GPT-5.6 Terra main judge · generation-agent harness and reasoning effort not stated on public leaderboard
CLIF BenchKimi Code · max effort · internal · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
CorpFin v2max effort · temp 1.0 · top_p 1.0 · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSearchQA · Atria paper scoreAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · in-house search/visit/python harness · 500 steps · 256K context / 64K max output · score subtype unstated · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSWE v1.15 results
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
mini-SWE / DeepSeek设置
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Kimi Code · max effort
mini-SWE-agent · official leaderboard snapshot
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
mini-swe-agent / Z.ai设置
E-Commerce Bench7 related benchmarks · 7 results
Mean final assets
Reported by E-Commerce Bench authors / Alibaba Qwen · via BenchBoard docs
benchmark-owner 365-day merchant simulation · five independent episodes · up to four stores · initial balance ¥100,000 · thinking enabled where supported; effort not specified · exact API snapshot not stated for unversioned labels
AnchorRatio · repeat-order pricing
Reported by E-Commerce Bench authors / Alibaba Qwen · via BenchBoard docs
benchmark-owner 365-day merchant simulation · five independent episodes · up to four stores · initial balance ¥100,000 · thinking enabled where supported; effort not specified · exact API snapshot not stated for unversioned labels
BadSpend · fraudulent-supplier spend
Reported by E-Commerce Bench authors / Alibaba Qwen · via BenchBoard docs
benchmark-owner 365-day merchant simulation · five independent episodes · up to four stores · initial balance ¥100,000 · thinking enabled where supported; effort not specified · exact API snapshot not stated for unversioned labels
Controllable return
Reported by E-Commerce Bench authors / Alibaba Qwen · via BenchBoard docs
benchmark-owner 365-day merchant simulation · five independent episodes · up to four stores · initial balance ¥100,000 · thinking enabled where supported; effort not specified · exact API snapshot not stated for unversioned labels
CSE+ · honest-deal negotiation
Reported by E-Commerce Bench authors / Alibaba Qwen · via BenchBoard docs
benchmark-owner 365-day merchant simulation · five independent episodes · up to four stores · initial balance ¥100,000 · thinking enabled where supported; effort not specified · exact API snapshot not stated for unversioned labels
Drawdown / peak assets
Reported by E-Commerce Bench authors / Alibaba Qwen · via BenchBoard docs
benchmark-owner 365-day merchant simulation · five independent episodes · up to four stores · initial balance ¥100,000 · thinking enabled where supported; effort not specified · exact API snapshot not stated for unversioned labels
Show 1 more resultHide 1 result
Yuan per tool call
Reported by E-Commerce Bench authors / Alibaba Qwen · via BenchBoard docs
benchmark-owner 365-day merchant simulation · five independent episodes · up to four stores · initial balance ¥100,000 · thinking enabled where supported; effort not specified · exact API snapshot not stated for unversioned labels
ExplorationBench22 related benchmarks · 22 results
AlienCode · M4 · fixed-probe control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
AlienCode · A4+O · complete rules after exploration
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · complete-rule reference or auxiliary diagnostic
AlienCode · M0 · baseline of Best@3 trajectory
Reported by Fudan University / Tencent Hy / Tsinghua University, Tencent Hy / ExplorationBench authors · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienCode · M4 · Best@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University, Tencent Hy / ExplorationBench authors · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienCode · M4 · direct answering control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
AlienCode · M4 · hindsight-replay control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
Show 16 more resultsHide 16 results
AlienCode · M4 · Mean@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University, Tencent Hy / ExplorationBench authors · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienCode · M4 · without-tool control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
AlienCode · M4 · Worst@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienCode · O@M0 · complete rules before exploration
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · complete-rule reference or auxiliary diagnostic
AlienCode · Rules correctly reported
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · complete-rule reference or auxiliary diagnostic
AlienLogic · A4+O · complete rules after exploration
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · complete-rule reference or auxiliary diagnostic
AlienLogic · Correct refusal on unprovable tasks
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · complete-rule reference or auxiliary diagnostic
AlienLogic · M0 · baseline of Best@3 trajectory
Reported by Fudan University / Tencent Hy / Tsinghua University, Tencent Hy / ExplorationBench authors · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienLogic · M4 · Best@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University, Tencent Hy / ExplorationBench authors · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienLogic · M4 · direct answering control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
AlienLogic · M4 · fixed-probe control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
AlienLogic · M4 · hindsight-replay control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
AlienLogic · M4 · Mean@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University, Tencent Hy / ExplorationBench authors · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienLogic · M4 · without-tool control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · one matched control trajectory
AlienLogic · M4 · Worst@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · 3 independent autonomous trajectories
AlienLogic · O@M0 · complete rules before exploration
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · max reasoning · complete-rule reference or auxiliary diagnostic
Faithfulness1 − hallucination rate · max effort · internal · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Finance BenchN/A harness · max effort · internal · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
KAETKimi Code · max effort · internal · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
KWV BenchN/A harness · max effort · internal · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
MIRA BenchMIRA harness · max effort · internal · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
MLE-Bench Lite · HumanRankAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · Claude Code · 12-hour task limit · HumanRank score · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Nejumi 428 related benchmarks · 28 results
ALT_バイアス
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
ALT_倫理・道徳
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
ALT_制御性
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
ALT_堅牢性
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
ALT_毒性
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
ALT_真実性
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
Show 22 more resultsHide 22 results
AVG_jaster_0shot
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
AVG_jaster_2shots
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_アプリケーション開発
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_コーディング
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_一般的知識
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_基礎的言語性能
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_専門的知識
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_応用的言語性能
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_情報検索
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_意味解析
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_抽象的推論
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_推論能力
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_数学的推論
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_構文解析
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_知識・質問応答
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_翻訳
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_表現
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_論理的推論
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
GLP_関数呼び出し
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
TOTAL_SCORE
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
アラインメント(ALT)_AVG
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
汎用的言語性能(GLP)_AVG
Source: Nejumi Leaderboard 4 data docs
moonshotai/kimi-k3: reasoning-max
NL2Repo-Bench3 results
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Setting not stated
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified · launch-page table
PerceptionBench9 related benchmarks · 10 results
Overall accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · four-run mean per paper Table 2 · GPT-oss-120B judge
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
three-run average · max effort · temp 1.0 · top_p .95
Attribute accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Comparison accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Contextual integration accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Counting accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Depth and 3D perception accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Show 3 more resultsHide 3 results
Fine-grained recognition accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Localization accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Visual relation accuracy
Reported by Moonshot AI / PerceptionBench authors · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
benchmark-owner evaluation · 3,000 verified questions total · open-ended short answers · unified prompts · max reasoning · main-table result; run count not specified · GPT-oss-120B judge
Reported by Z.ai · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified
RecreationBench paper26 related benchmarks · 26 results
Web · Visual assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Web · nominal 50-task split · platform-specific score
Android · Applications with Prog =100%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Android · nominal 50-task split · platform-specific score
Android · Applications with Prog ≥90%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Android · nominal 50-task split · platform-specific score
Android · Programmatic assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Android · nominal 50-task split · platform-specific score
Android · Visual assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Android · nominal 50-task split · platform-specific score
macOS · Applications with Prog =100%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · macOS · nominal 50-task split · platform-specific score
Show 20 more resultsHide 20 results
macOS · Applications with Prog ≥90%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · macOS · nominal 50-task split · platform-specific score
macOS · Programmatic assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · macOS · nominal 50-task split · platform-specific score
macOS · Visual assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · macOS · nominal 50-task split · platform-specific score
Overall · Applications with Prog =100%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Overall · 250 held-out tasks · equal-weight platform macro average
Overall · Applications with Prog ≥90%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Overall · 250 held-out tasks · equal-weight platform macro average
Overall · Estimated API cost per task
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · estimated mean API cost per task · 90% cache reads assumed
Overall · Programmatic + visual mean
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Overall · 250 held-out tasks · equal-weight platform macro average
Overall · Programmatic assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Overall · 250 held-out tasks · equal-weight platform macro average
Overall · Visual assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Overall · 250 held-out tasks · equal-weight platform macro average
Ubuntu · Applications with Prog =100%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Ubuntu · nominal 50-task split · platform-specific score
Ubuntu · Applications with Prog ≥90%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Ubuntu · nominal 50-task split · platform-specific score
Ubuntu · Programmatic assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Ubuntu · nominal 50-task split · platform-specific score
Ubuntu · Visual assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Ubuntu · nominal 50-task split · platform-specific score
Web · Applications with Prog =100%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Web · nominal 50-task split · platform-specific score
Web · Applications with Prog ≥90%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Web · nominal 50-task split · platform-specific score
Web · Programmatic assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Web · nominal 50-task split · platform-specific score
Windows · Applications with Prog =100%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Windows · nominal 50-task split · platform-specific score
Windows · Applications with Prog ≥90%
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Windows · nominal 50-task split · platform-specific score
Windows · Programmatic assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Windows · nominal 50-task split · platform-specific score
Windows · Visual assertion score
Reported by QwenLM / RecreationWorld authors · via BenchBoard docs
RecreationWorld benchmark-owner paper run · 250 held-out tasks · GUI exploration + coding tools + visual verification · max reasoning effort per site data · 1m context · 128k max output tokens · Windows · nominal 50-task split · platform-specific score
SaaS-Benchmax effort · temp 1.0 · top_p 1.0 · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
SkillsBench · Text-only (79 tasks)Hy4 official Benchmark Appendix · highest available reasoning setting · Claude Code · 79 self-contained text-only tasks · three-run average · Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Swarm BenchKimi Agent · max effort · internal · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Vals21 related benchmarks · 23 results
Show 17 more resultsHide 17 results
Immigration
Source: Vals AI
LegalBench
Source: Vals AI
MedCode
Source: Clinical Benchmarks data docs · Sep 2026
Setting not stated
Source: Vals AI
Setting not stated
MedScribe
Source: Clinical Benchmarks data docs · Sep 2026
Setting not stated
Source: Vals AI
Setting not stated
MortgageTax
Source: Vals AI
SAGE
Source: Vals AI
TaxEval v2
Source: Vals AI
WideSearch · Atria paper scoreAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · in-house search/visit/python harness · 500 steps · 256K context / 64K max output · score subtype unstated · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
WildClawBench6 related benchmarks · 6 results
WildClawBench
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Multimodal (26 tasks)
Reported by WildClawBench / InternLM · via BenchBoard docs
OpenClaw · multimodal-input split · 26 tasks
Overall
Reported by WildClawBench / InternLM · via BenchBoard docs
OpenClaw · full 60-task suite
Pure-text (34 tasks)
Reported by WildClawBench / InternLM · via BenchBoard docs
OpenClaw · pure-text split · 34 tasks
Total Cost
Reported by WildClawBench / InternLM · via BenchBoard docs
OpenClaw · full 60-task suite
Total Elapsed
Reported by WildClawBench / InternLM · via BenchBoard docs
OpenClaw · full 60-task suite
Workspace-Bench · English Full · Atria runAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · AgentCompass · English Full · two-hour limit · 2 CPUs / 8 GiB · Seed 2.0 Lite judge · GPT in Codex, others in Claude Code · metric subtype unstated · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Workspace-Bench · English Lite · Atria runAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · AgentCompass · English Lite · two-hour limit · 2 CPUs / 8 GiB · Seed 2.0 Lite judge · GPT in Codex, others in Claude Code · metric subtype unstated · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Reported by DeepSeek · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
DeepSeek V4.1 launch chart · settings unspecified · with tools
Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
pass@5 · max effort
τ³-Bankingmax effort · temp 1.0 · top_p 1.0 · Reported by Moonshot AI · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
τ³-Bench · Banking Knowledge · Atria runAtria Dawn Preview official card / report Table 1 · published comparison run · model snapshots and reasoning tiers not specified unless stated · 97 Banking Knowledge tasks · modified Tau2/Tau3 runner from commit a2c0247 · alltools BM25/dense/shell retrieval · GPT-5.2 low judge and simulator · 200-step limit · Reported by Shanghai AI Laboratory · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Hugging Face official benchmark leaderboards; licence varies by benchmark · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · Clinical Benchmarks (clinicalbenchmarks.ai), CC BY 4.0; fractions converted to percentages; source documents retain their own terms (licence) · LMArena, CC BY 4.0 (licence) · Data by LLM Stats · Baromètre de l’écriture IA, Johannes Eva; public numeric ranking data, © 2026 Johannes Eva, no results licence stated · BullshitBench by Peter Gostev, MIT (licence) · EQ-Bench (Samuel J. Paech) — cite paech2023eqbench · Keelemudelite mõõdupuu: Independent LLM leaderboard for Estonian. Eesti Keele Instituut. https://moodupuu.eki.ee, CC BY 4.0. Latest per-benchmark results selected with partial-run flags retained. (licence) · OpenRouter public benchmark leaderboard and search-lane exports, CC BY 4.0. Fractional source scores converted to percentages; immutable dated snapshots. (licence) · MCPMark (eval-sys), mcpmark.ai Verified leaderboard. Single-run Pass@1 fractions shown as percentages. · MineBench, MIT (licence) · Nejumi Leaderboard 4, Weights & Biases Japan; public publisher-selected numeric evaluation results · Outback Yak Research, NepNewsBench v2 research draft. Public numerical aggregates from a candidate preview that the publisher says is not evidence of publication; no release licence assigned and none asserted here. · Sierra Research tau2-bench leaderboard submissions (github.com/sierra-research/tau2-bench), MIT (licence) · WeirdML v3 by Håvard Tveit Ihle (Norwegian Defence Research Establishment). API costs supported by Epoch AI, METR and NDRE.

Also known as: accounts/fireworks/models/kimi-k3, accounts/fireworks/routers/kimi-k3-fast, aihubmix/coding-kimi-k3, aihubmix/kimi-k3, alibaba/kimi-k3, aws-bedrock/kimi-k3, azure_ai/fw-kimi-k3, baseten/moonshotai/kimi-k3, bedrock/us-gov-east-1/moonshotai.kimi-k3, bedrock/us-gov-west-1/moonshotai.kimi-k3, cline-pass/kimi-k3, coding-kimi-k3, coding-kimi-k3-free, databricks/databricks-kimi-k3, deepinfra/moonshotai/kimi-k3, e2ee-kimi-k3-p, empiriolabs/kimi-k3, fireworks/kimi-k3, fireworks/kimi-k3-fast, fireworks_ai/accounts/fireworks/models/kimi-k3, fireworks_ai/accounts/fireworks/routers/kimi-k3-fast, fireworks_ai/kimi-k3, fireworks_ai/kimi-k3-fast, github_copilot/kimi-k3, global.moonshotai.kimi-k3, gmi/moonshotai/kimi-k3, hf:moonshotai/kimi-k3, kimi k3, kimi k3 fast, kimi-k3-fast, kimi-k3-fast-api, kimi-k3-max, kimi-k3:fast, kimi-k3@eu, kimi/kimi-k3, moonshot-ai/kimi-k3, moonshot/kimi-k3, moonshotai.kimi-k3, moonshotai/kimi-k3, moonshotai/kimi-k3-fast, moonshotai/kimi-k3-free, moonshotai/kimi-k3-tee, moonshotai/kimi-k3:batch, nebius/kimi-k3, nebius/moonshotai/kimi-k3, novita/kimi-k3, novita/moonshotai/kimi-k3, opencode_go/kimi-k3, parasail-kimi-k3, parasail-kimi-k3-fast, perplexity/kimi-k3, runpod/kimi-k3, runware/kimi-k3, scx-ai-gp/kimi-k3, scx-ai/kimi-k3, tee/kimi-k3, tencent/kimi-k3, tensorx/kimi-k3, together-ai/kimi-k3, together_ai/moonshotai/kimi-k3, umans-kimi-k3, us.moonshotai.kimi-k3

Like this one. Tap or hover any dotted value, label or icon to see where it comes from and how it is measured. open details when tapped or hovered. How this data works