Skip to content

Tencent · Chat

Hy4 preview

Tencent 🇨🇳ChatHy4 familyOpen weights. One of its sources says the weights are publicly available.
From Neosantara's listing:
Tencent Hunyuan 4 Preview. Large-scale MoE reasoning model with 1M context window, optimized for complex reasoning, programming, and long-horizon tool execution.
llminfo IQ
136.6
Max context
1M
Listings
54

Evidence and sources

🤗 tencent/Hy4-preview780B params Model architecture, as its sources state it llminfo IQ 136.6, ranked among chat models. From 95 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 118.2, 95% interval 102–135. From 15 benchmark families across 4 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 (56.4). Cheaper per task than its benchmark results predict. From 29 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 resultsChatty (57 ±9). Generates more tokens than models with similar coding results. Measured at max effort. Little data behind it, so treat it as a rough estimate.Median 309,296 output tokens per task at max effort.How the Yap score worksReleased 28 Aug 2026Maker HQ: CN (according to Tencent's own website)
All rankings: #43 on OpenRouter this month #113 on OpenRouter this week #141 on OpenRouter today #244 on NanoGPT recentlySources: LMArena data, Models.dev data, AI/ML API data, Morpheus Inference Marketplace data, Neosantara Inference data, llminfo researchPrimary data from LMArena from Models.dev: weights availability from AI/ML API: primary mode from Morpheus Inference Marketplace: architecture, parameter count from Neosantara Inference: description from llminfo research: release date

Advantages and trade-offs

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

Advantages

Open weights available

The weights are public. 3 of the 9 peers that say either way have open weights.

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
Text

Capabilities

Supported operations and provider-stated features

ReasoningFunction callingStructured outputPrompt cachingWeb search

Listings

 Tencenttencent/hy4-preview · fp8Source: LiteLLM data960K$0.417cache read $0.021$1.2505
 Vancinehy4-previewSource: Models.dev data1M$0.67cache read $0.034$2.00
 WattHy4-previewSource: Models.dev data1M$0.83$2.50
 Atlascloudtencent/hy4-preview · fp8Source: AtlasCloud data1M$0.834cache read $0.042$2.501
 Eden AIInference by DeepInfradeepinfra/tencent/Hy4-previewSource: Models.dev data1M$0.834cache read $0.042$2.501
 Fastroutertencent/hy4-previewSource: FastRouter data1M$0.834cache read $0.042$2.501
 Flowdottencent/Hy4-previewSource: FlowDot data1M$0.834$2.501
 Hugging FaceInference by DeepInfratencent/Hy4-previewSource: Hugging Face Inference Providers data1M$0.834$2.501
 Kilotencent/hy4-previewSources: Kilo AI Gateway data, Models.dev data1M$0.834cache read $0.042$2.501
 LlmgatewayInference by DeepInfratencent/Hy4-previewSource: LLMGateway dataNot published by the provider$0.834cache read $0.042$2.501
Show all 49 listings

Showing 10 of 49: the maker's own listing first, then the cheapest per 1M input tokens from each provider.

Also sold by Aihubmix, Aimlapi, Commandcode, DeepInfra, Gmicloud, Infersia, Ionet, Llmgateway Providers, NanoGPT, Neosantara, Netmind, Nous Portal, Novita AI, Opencode Go, OpenRouter, Ppio, Qiniu, Venice, Vercel, Wavespeed, Zenmux.

5 listings without a price are hidden.

Included in plans

Benchmarks

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

Agentic

Apex Agents2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Arena AgentSource: LMArena data
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Draco2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
GDPval-AASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
GDPval-AA v2official model-card appendix · Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
JobBench2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
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 Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SWE-Bench Pro4 related benchmarks · 5 results
SWE-Bench Pro
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: Hugging Face Official Benchmarks data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SWE Atlas - Test Writing
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
SWE-Atlas · Codebase Q&A
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
SWE-Atlas · Refactoring
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: OpenRouter benchmark exports
27 Aug 2026 version · 90-day results
Across 11 providers lowest 72.7% median 76% highest 77.6%OpenRouter's automatic routing 76%
Show results from 11 providers and OpenRouter's automatic routingHide results from 11 providers and OpenRouter's automatic routing
  • DeepInfra
  • Makora
  • Venice
  • Tencent
  • Novita
  • OpenRouter's automatic routing
  • Cohere
  • GMICloud
  • AtlasCloud
  • SiliconFlow
  • InferenceNet
  • Io Net

Reasoning

Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Humanity's Last Exam4 related benchmarks · 4 results
Humanity's Last Exam
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
text-only subset · no tools
(no tools, text-only)
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
(with tools, text-only)
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
(with tools)
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
text-only subset · with tools
OfficeQA Pro2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
SuperChem2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
WideSearch2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Coding

CyberGym2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
DeepSWESource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
NL2RepoSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
SWE-Marathon2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
ValsSource: Vals AI

Science

CritPT2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
GPQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
GPQA Diamond13 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: OpenRouter benchmark exports
27 Aug 2026 version · 90-day results
Across 11 providers lowest 86% median 89.5% highest 90.7%OpenRouter's automatic routing 89.7%
Show results from 11 providers and OpenRouter's automatic routingHide results from 11 providers and OpenRouter's automatic routing
  • DeepInfra
  • InferenceNet
  • Cohere
  • SiliconFlow
  • OpenRouter's automatic routing
  • GMICloud
  • Tencent
  • Venice
  • Makora
  • AtlasCloud
  • Novita
  • Io Net

Math

ArxivMath2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official model-card appendix
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
HorizonMath2 results
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
Setting not stated

Multilingual

EKI EstonianSource: Keelemudelite mõõdupuu data docs · Sep 2026

General

Human preference

Other

BrokenArXivofficial model-card appendix · Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
E-BenchReported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
ExplorationBench22 related benchmarks · 22 results
AlienCode · M4 · fixed-probe control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · one matched control trajectory
AlienCode · A4+O · complete rules after exploration
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high 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 · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high 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 · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · 3 independent autonomous trajectories
AlienCode · M4 · direct answering control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · one matched control trajectory
AlienCode · M4 · hindsight-replay control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high 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 · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · 3 independent autonomous trajectories
AlienCode · M4 · without-tool control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · one matched control trajectory
AlienCode · M4 · Worst@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · 3 independent autonomous trajectories
AlienCode · O@M0 · complete rules before exploration
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · complete-rule reference or auxiliary diagnostic
AlienCode · Rules correctly reported
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high 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 · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · complete-rule reference or auxiliary diagnostic
AlienLogic · Correct refusal on unprovable tasks
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high 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 · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high 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 · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · 3 independent autonomous trajectories
AlienLogic · M4 · direct answering control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · one matched control trajectory
AlienLogic · M4 · fixed-probe control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · one matched control trajectory
AlienLogic · M4 · hindsight-replay control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high 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 · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · 3 independent autonomous trajectories
AlienLogic · M4 · without-tool control
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · one matched control trajectory
AlienLogic · M4 · Worst@3 held-out accuracy
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · 3 independent autonomous trajectories
AlienLogic · O@M0 · complete rules before exploration
Reported by Fudan University / Tencent Hy / Tsinghua University · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
official 2026-10-13 leaderboard snapshot · 70 held-out tasks in this world · 4 exploration rounds · 3 answers/task · high reasoning · complete-rule reference or auxiliary diagnostic
Harbor Indexofficial model-card appendix · Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Hy-LifeSearchReported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
Hy47 related benchmarks · 7 results
vs GLM-5.3 · Tie
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
163 internal experts · 203 engineering tasks · blind side-by-side evaluation
Expert Blind Eval · Average
Reported by Tencent Hy, Tencent · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
163 internal experts · 203 engineering tasks · blind side-by-side evaluation
vs GLM-5.3 · Loss
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
163 internal experts · 203 engineering tasks · blind side-by-side evaluation
vs GLM-5.3 · Win
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
163 internal experts · 203 engineering tasks · blind side-by-side evaluation
vs Kimi K3 · Loss
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
163 internal experts · 203 engineering tasks · blind side-by-side evaluation
vs Kimi K3 · Tie
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
163 internal experts · 203 engineering tasks · blind side-by-side evaluation
Show 1 more resultHide 1 result
vs Kimi K3 · Win
Reported by Tencent Hy · via BenchBoard docs · Reported by the model's developer. The source relaying it did not verify it.
163 internal experts · 203 engineering tasks · blind side-by-side evaluation
NL2Repo-Benchofficial model-card appendix · Reported by Tencent Hy · 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.
SuperCLUESource: SuperCLUE data
Vals13 related benchmarks · 15 results
EMB
Source: Vals AI
EMB v1.1
Source: Vals AI
Show 9 more resultsHide 9 results
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
Router Index
Source: Vals AI
SRE Bench
Source: Vals AI

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. · Data by LLM Stats · LMArena, CC BY 4.0 (licence) · Baromètre de l’écriture IA, Johannes Eva; public numeric ranking data, © 2026 Johannes Eva, no results licence stated · Clinical Benchmarks (clinicalbenchmarks.ai), CC BY 4.0; fractions converted to percentages; source documents retain their own terms (licence) · 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) · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · OpenRouter public benchmark leaderboard and search-lane exports, CC BY 4.0. Fractional source scores converted to percentages; immutable dated snapshots. (licence) · Hugging Face official benchmark leaderboards; licence varies by benchmark

Also known as: aihubmix/hy4-preview, deepinfra/hy4-preview, deepinfra/tencent/hy4-preview, gmi/tencent/hy4-preview, hy4 preview, novita/hy4-preview, opencode_go/hy4-preview, tencent/hy4-preview

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