Skip to content

Alibaba · Chat

Qwen3.5 4B

Alibaba 🇨🇳ChatQwen familyOpen weights. One of its sources says the weights are publicly available.
From Empiriolabs's listing:
Qwen3.5 4B is a low-cost multimodal reasoning model with 256K context, image and video input, function tools, and structured output.
llminfo IQ
107.0
Market median price
$0.0475 blended / 1M
Max context
262K
Listings
6

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/Qwen3.5-4Bllminfo IQ 107.0, ranked among chat models. From 129 benchmarks. Plenty of data behind it.How llminfo IQ is computedCoding IQ 77.8, 95% interval 66–90. From 3 benchmark families across 3 sources. Little data behind it, so treat it as a rough estimate. Uses the model's best published results. It has its own scale, separate from llminfo IQ.How Coding IQ is computedReleased 27 Feb 2026Maker HQ: CN (source)
Sources: EmpirioLabs data, Models.dev data, Hugging FacePrimary data from EmpirioLabs from Models.dev: weights availability from Hugging Face: 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. 2 of the 8 peers that say either way have open weights.

Trade-offs

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

 Empiriolabsqwen3-5-4b · FP8 weights + FP8 KVSources: EmpirioLabs data, LiteLLM data262K$0.04cache read $0.02$0.07
 Featherless AIQwen/Qwen3.5-4BSource: Featherless data33K$0.10$0.20
 Merge Gatewayqwen/qwen3.5-4bSource: Models.dev data262K$0.04$0.07
 Thinking MachinesQwen/Qwen3.5-4BSource: Models.dev data66K$0.33cache read $0.066$1.005

2 listings without a price are hidden.

Benchmarks

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

Vision

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

OCR

OmniDocBenchSource: Nanonets IDP data

Math

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

Science

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

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

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

Coding

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

Long context

Agentic

BFCL-V4Source: LLM Stats data · Reported by the model's developer. The source relaying it did not verify it. claim

Human preference

Other

Nejumi 428 related benchmarks · 28 results
ALT_バイアス
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
ALT_倫理・道徳
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
ALT_制御性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
ALT_堅牢性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
ALT_毒性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
ALT_真実性
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
Show 22 more resultsHide 22 results
AVG_jaster_0shot
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
AVG_jaster_2shots
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_アプリケーション開発
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_コーディング
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_一般的知識
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_基礎的言語性能
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_専門的知識
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_応用的言語性能
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_情報検索
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_意味解析
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_抽象的推論
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_推論能力
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_数学的推論
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_構文解析
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_知識・質問応答
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_翻訳
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_表現
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_論理的推論
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
GLP_関数呼び出し
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
TOTAL_SCORE
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
アラインメント(ALT)_AVG
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: reasoning-enabled
汎用的言語性能(GLP)_AVG
Source: Nejumi Leaderboard 4 data docs
Qwen/Qwen3.5-4B: 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). · Data by LLM Stats · Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · EQ-Bench (Samuel J. Paech) — cite paech2023eqbench · OpenEvals/leaderboard-data, MIT (licence) · 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). · 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 · 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)

Around the web

  • Alibaba unveils advanced Qwen 3 AI as Chinese tech rivalry intensifies | Reuters
    By Reuters April 30, 2025 6:57 PM UTC Updated April 30, 2025 April 29 (Reuters) …By Reuters April 30, 2025 6:57 PM UTC Updated April 30, 2025 April 29 (Reuters) - Chinese tech giant Alibaba Group (9988.HK) , opens new tab launched Qwen 3 on Tuesday, an upgraded version of its fla…

    reuters.com

  • Qwen 3: The new open standard - by Nathan Lambert
    and modern post-training has enabled crazy hill climbing on the evals that are currently in …and modern post-training has enabled crazy hill climbing on the evals that are currently in vogue. In an ICLR workshop panel on open models , Junyang Lin from Qwen said roughly that they “need 100 pe…

    interconnects.ai

  • Qwen 3 Max (Alibaba) - AI
    title: Qwen 3 Max description: Alibaba's Qwen 3 Max is a large language model with …title: Qwen 3 Max description: Alibaba's Qwen 3 Max is a large language model with strong coding, reasoning, and multilingual capabilities, served via DashScope's OpenAI-compatible endpoint. image: h…

    developers.cloudflare.com

  • Qwen 3 Models — Complete Guide Including Qwen3-Next (2026)
    across the board Qwen 3.5 is out. The Qwen 3.5 generation (February 2026) surpasses Qwen …across the board Qwen 3.5 is out. The Qwen 3.5 generation (February 2026) surpasses Qwen 3 on nearly every benchmark — with native multimodal support, faster inference, and a 35B-A3B model that outpe…

    qwen-ai.com

  • Qwen AI for Everyone | Qwen 3, Qwen Studio & Model
    expose port 8000. Add --gpus all for NVIDIA GPU support. Copy $ docker run -d …expose port 8000. Add --gpus all for NVIDIA GPU support. Copy $ docker run -d \ --name qwen \ --gpus all \ -p 8000:8000 \ -v qwen-models:/root/.cache/qwen \ qwenlm/qwen-server:latest \ --model qwen-3…

    qwen3lm.com

  • Qwen 3 32B by Alibaba on Vercel AI Gateway, Specs, Pricing & API
    use cases. Sessions that exceed the window will require context management strategies like summarizing earlier …use cases. Sessions that exceed the window will require context management strategies like summarizing earlier turns. What tool-calling capabilities does Qwen 3 32B have? Qwen 3 32B supports tool cal…

    vercel.com

  • Qwen 3 32B - GroqDocs - console.groq.com
    Get Started with Qwen 3 32B Experience state-of-the-art language understanding and generation with Qwen 3 …Get Started with Qwen 3 32B Experience state-of-the-art language understanding and generation with Qwen 3 32B with Groq speed: curl JavaScript Python JSON shell pip install groq Python from groq impo…

    console.groq.com

  • Alibaba to release its flagship Qwen models this month | Cybernews
    Several Qwen 3 models are to be released this April, according to the media. Cloud …Several Qwen 3 models are to be released this April, according to the media. Cloud computing company Alibaba Cloud, a subsidiary of the Chinese conglomerate Alibaba Group, is preparing to launch its …

    cybernews.com

  • Ultimate Guide - The Best Qwen Models in 2026
    Which Qwen models made it into our top three picks for 2026? Our top three …Which Qwen models made it into our top three picks for 2026? Our top three Qwen models for 2026 are Qwen3-235B-A22B (the flagship general-purpose model), Qwen3-Coder-480B-A35B-Instruct (the advanced …

    siliconflow.com

  • Qwen 3 (235B) - Vals AI
    Code 6/9/2026 Anthropic Claude Fable 5 6/4/2026 NVIDIA Nemotron 3 Ultra 6/1/2026 Alibaba Qwen 3.7 …Code 6/9/2026 Anthropic Claude Fable 5 6/4/2026 NVIDIA Nemotron 3 Ultra 6/1/2026 Alibaba Qwen 3.7 Plus 5/31/2026 MiniMax MiniMax-M3 5/28/2026 Anthropic Claude Opus 4.8 5/22/2026 Devin SWE-1.6 Fast 5/…

    vals.ai

Also known as: empiriolabs/qwen3-5-4b, qwen/qwen3.5-4b, qwen3-5-4b

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