Skip to content

Alibaba · Chat

Qwen3.5 9B

Alibaba 🇨🇳ChatQwen familyOpen weights. One of its sources says the weights are publicly available.
From Empiriolabs's listing:
Qwen3.5 9B is a compact multimodal reasoning model with 256K context, image and video input, function tools, and structured output.
llminfo IQ
113.2
Market median price
$0.1313 blended / 1M
Max context
262K
Listings
59

Evidence and sources

International storefront (alibabacloud.com) is Singapore-incorporated; mainland operating entity 杭州阿里云智能科技有限公司 (Hangzhou).

🤗 Qwen/Qwen3.5-9Bllminfo IQ 113.2, ranked among chat models. From 223 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 85.6, 95% interval 77–94. From 6 benchmark families across 5 sources. A fair amount 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 (55.4). Cheaper per task than its benchmark results predict. From 10 benchmarks with measured task costs. Plenty of data behind it.How Value for money worksReleased 27 Feb 2026Maker HQ: CN (source)
All rankings: #12 on NanoGPT recently #97 on OpenRouter this month #103 on OpenRouter this week #130 on OpenRouter todaySources: Darkbloom data docs, DeepInfra data, EmpirioLabs data, flex.ai API data, Models.dev data, Hugging FacePrimary data from Darkbloom from DeepInfra: name, family, variant from EmpirioLabs: primary mode, description from flex.ai API: hugging face id from Models.dev: weights availability from Hugging Face: release date
Tags from its sources:instructreasoningtoolsvisionopen-sourcelong-contextefficient

Advantages and trade-offs

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

Advantages

Open weights available

The weights are public. 4 of the 8 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, 06:07 UTC.Prices come from the model's maker where they sell it, otherwise their official partner, otherwise the market median. This comparison uses 6 maker's prices; 4 market median prices.

Input & output

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

Capabilities

Supported operations and provider-stated features

VisionReasoningFunction callingStructured outputPrompt caching

Listings

 NanoGPTInference by GreenferenceRegions: 🇪🇺 EUqwen/qwen3.5-9b · fp8Source: NanoGPTNot published by the provider$0.042cache read $0.021$0.0683140.2252ms
 Siliconflow CnQwen/Qwen3.5-9BSource: Models.dev data262K$0.0747$0.5978——
 OpenRouterInference by Darkbloomqwen/qwen3.5-9b · fp4Source: OpenRouter data docs262K$0.08cache read $0.04$0.13971,071.5ms
 DarkbloomQwen3.5-9B · fp4Source: Darkbloom data docsNot published by the provider$0.08cache read $0.04$0.13——
 DoublewordQwen/Qwen3.5-9BSource: GenAI Prices dataNot published by the provider$0.08$0.70——
 Empiriolabsqwen3-5-9b · FP8 weights + FP8 KVSources: EmpirioLabs data, LiteLLM data262K$0.09cache read $0.045$0.13——
 Merge Gatewayqwen/qwen3.5-9bSource: Models.dev data262K$0.09$0.13——
 Fastrouterqwen/qwen3.5-9bSource: FastRouter data262K$0.10$0.15——
 FlowdotQwen/Qwen3.5-9BSource: FlowDot data262K$0.10$0.15——
 Hugging FaceInference by DeepInfraQwen/Qwen3.5-9BSource: Hugging Face Inference Providers data262K$0.10$0.15——

9 listings without a price are hidden.

Benchmarks

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

Multilingual

IdrockBench Uzbek4 related benchmarks · 4 results
MMMLUSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
WMT24++Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Science

GPQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
GPQA Diamond10 results
Source: OpenEvals Leaderboard data · Reported by the model's developer. The source relaying it did not verify it. · Mar 2026
Setting not stated
Source: OpenRouter benchmark exports
10 Mar 2026 version · 90-day results
Across 7 providers lowest 70.2% median 79.5% highest 81%OpenRouter's automatic routing 77.4%
Show results from 7 providers and OpenRouter's automatic routingHide results from 7 providers and OpenRouter's automatic routing
  • PrimeIntellect
  • Parasail
  • DeepInfra
  • Together
  • Venice
  • OpenRouter's automatic routing
  • SiliconFlow
  • Darkbloom
SuperGPQASource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Vision

Math

HMMT 2025Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
HMMT25Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
MathArenaSource: MathArena data
PolyMATHSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

OCR

OmniDocBenchSource: Nanonets IDP data

Agentic

BFCL-V4Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
GAIA4 results
Source: GAIA Leaderboard data · May 2026
Difficulty level
Difficulty not reported
Source: OpenRouter benchmark exports
10 Mar 2026 version · 90-day results
Across 6 providers lowest 66.2% median 70.4% highest 76.2%OpenRouter's automatic routing 75%
Show results from 6 providers and OpenRouter's automatic routingHide results from 6 providers and OpenRouter's automatic routing
  • DeepInfra
  • Together
  • OpenRouter's automatic routing
  • Darkbloom
  • PrimeIntellect
  • SiliconFlow
  • Venice

Long context

Reasoning

C-EvalSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
t2-benchSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Coding

Instruction following

IFBenchSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
IFEvalSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Knowledge

IncludeSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
MMLU-ProSource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

General

MAXIFESource: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim
NOVA-63Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Human preference

Other

ArmBench-LLM Armenian22 related benchmarks · 22 results
Horangi36 related benchmarks · 36 results
Nejumi 428 related benchmarks · 28 results
ALT_バイアス
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
ALT_倫理・道徳
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
ALT_制御性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
ALT_堅牢性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
ALT_毒性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
ALT_真実性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
Show 22 more resultsHide 22 results
AVG_jaster_0shot
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
AVG_jaster_2shots
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_アプリケーション開発
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_コーディング
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_一般的知識
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_基礎的言語性能
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_専門的知識
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_応用的言語性能
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_情報検索
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_意味解析
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_抽象的推論
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_推論能力
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_数学的推論
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_構文解析
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_知識・質問応答
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_翻訳
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_表現
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_論理的推論
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
GLP_関数呼び出し
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
TOTAL_SCORE
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
アラインメント(ALT)_AVG
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled
汎用的言語性能(GLP)_AVG
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-9B: reasoning-enabled

Data credits: SEA-HELM leaderboard, AI Singapore (AI Products team), National University of Singapore. Published aggregate scores; results licence not stated (SEA-HELM code is MIT). · ArmBench-LLM by Metric-AI. Public numeric primary task results; no separate results licence asserted. · Data by LLM Stats · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · OpenEvals/leaderboard-data, MIT (licence) · OpenRouter public benchmark leaderboard and search-lane exports, CC BY 4.0. Fractional source scores converted to percentages; immutable dated snapshots. (licence) · Horangi 4, Weights & Biases Korea; public publisher-selected numeric evaluation results · IdrockBench, IDROCK AI Excellence Center, New Uzbekistan University. Public numeric results; no results licence stated (code MIT and dataset licences not borrowed). · JMLE2026-Bench and IgakuQA119 by Naoto Iwase; results licence not stated (code MIT, question data CC BY 4.0). · EQ-Bench (Samuel J. Paech) — cite paech2023eqbench · LLM Chess by Maxim Saplin, Apache-2.0 (Copyright [2024] [Maxim Saplin]) (licence) · Hugging Face official benchmark leaderboards; licence varies by benchmark · MathArena (ETH SRI / INSAIT), MIT — cite dekoninck2026matharena (licence) · Nejumi Leaderboard 4, Weights & Biases Japan; public publisher-selected numeric evaluation results · Shaberi v2.1, Shisa.AI, Apache-2.0. GPT-5.1-judge scores only; judges are not comparable. (licence) · Swallow LLM Leaderboard v2, Swallow LLM team (Institute of Science Tokyo). Contents CC BY 4.0; Japanese post-trained task scores on the source's 0..1 scale. (licence) · UzLiB, Tahrirchi (Abror Shopulatov). Public LEADERBOARD.md numbers; repository MIT covers software and documentation, no separate results licence stated and none asserted.

Around the web

  • Qwen3 - How to Run & Fine-tune | Unsloth Documentation
    We also uploaded Qwen3 with native 128K context length. Qwen achieves this by using YaRN …We also uploaded Qwen3 with native 128K context length. Qwen achieves this by using YaRN to extend its original 40K window to 128K. Unsloth also now supports fine-tuning and Reinforcement Learning (R…

    unsloth.ai

    Docs
  • [2505.09388] Qwen3 Technical Report
    > Abstract:In this work, we present Qwen3, the latest version of the Qwen model family. …> Abstract:In this work, we present Qwen3, the latest version of the Qwen model family. Qwen3 comprises a series of large language models (LLMs) designed to advance performance, efficiency, and multi…

    arxiv.org

    Paper
  • GitHub - QwenLM/Qwen3: Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud. · GitHub
    Previous Qwen3 Release ### Qwen3 (aka Qwen3-2504) We are excited to announce the release of …Previous Qwen3 Release ### Qwen3 (aka Qwen3-2504) We are excited to announce the release of Qwen3, the latest addition to the Qwen family of large language models. These models represent our most adv…

    github.com

    Repository
  • Qwen3 - a Qwen Collection
    #### Qwen/Qwen3-30B-A3B-Base #### Qwen/Qwen3-14B-Base #### Qwen/Qwen3-8B-Base #### Qwen/Qwen3-4B-Base #### Qwen/Qwen3-1.7B-Base #### Qwen/Qwen3-0.6B-Base #### Qwen/Qwen3-4B-MLX-8bit #### …#### Qwen/Qwen3-30B-A3B-Base #### Qwen/Qwen3-14B-Base #### Qwen/Qwen3-8B-Base #### Qwen/Qwen3-4B-Base #### Qwen/Qwen3-1.7B-Base #### Qwen/Qwen3-0.6B-Base #### Qwen/Qwen3-4B-MLX-8bit #### Qwen/Qwen3-4…

    huggingface.co

    Model card
  • Qwen (Qwen)

    #### Qwen/Qwen3.8-27B-FP8 #### Qwen/Qwen3.8-Flash-Next #### Qwen/Qwen3.8-Flash-Next-FP8 #### Qwen/Qwen3.8-2.4T-A95B #### Qwen/Qwen3.8-2.4T-A95B-FP8 #### Qwen/Qwen3.8-27B #### Qwen/Qwen3.8-27B-FP8 ###…

    huggingface.co

    Model card
  • What's the deal with Qwen3 ? Why does it seem like ...
    Qwen 3 are very capable vision-language models, with a fairly standard decoder-only transformer architecture that …Qwen 3 are very capable vision-language models, with a fairly standard decoder-only transformer architecture that is well suited for working as

    reddit.com

    Discussion
  • qwen3
    The Qwen 3 family is a comprehensive suite of dense and mixture-of-experts (MoE) models. The …The Qwen 3 family is a comprehensive suite of dense and mixture-of-experts (MoE) models. The flagship model, Qwen3-235B-A22B, achieves competitive results in benchmark evaluations of coding, math, ge…

    ollama.com

  • Qwen-3: Alibaba Cloud's Next-Gen Open Source LLM | Apache 2.0 | MoE & Dense
    Undergoes three-stage pre-training based on nearly 36 trillion tokens of data, and employs four-stage post-training …Undergoes three-stage pre-training based on nearly 36 trillion tokens of data, and employs four-stage post-training to develop hybrid thinking and general capabilities. · 36T tokens pre-training data …

    qwen-3.com

  • Qwen3: Think Deeper, Act Faster | Qwen
    Today, we are excited to announce the release of Qwen3, the latest addition to the …Today, we are excited to announce the release of Qwen3, the latest addition to the Qwen family of large language models. Our flagship model, Qwen3-235B-A22B, achieves competitive results in benchmark…

    qwenlm.github.io

  • Qwen3: Features, DeepSeek-R1 Comparison, Access, and More | DataCamp
    In simple terms: the big models were trained first, and then the lightweight ones were …In simple terms: the big models were trained first, and then the lightweight ones were distilled from them. This way, the full Qwen3 family shares a similar style of thinking, even across very differ…

    datacamp.com

Also known as: deepinfra/qwen/qwen3.5-9b, deepinfra/qwen3.5-9b, empiriolabs/qwen3-5-9b, parasail-qwen35-9b, qwen/qwen3.5-9b, qwen/qwen3.5-9b-fp8, qwen3-5-9b, qwen3_5-9b-mlx-4bit, qwen3_5-9b-q4_k_m, together_ai/qwen/qwen3.5-9b

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