Inference
EN
Alibaba

Qwen3.5-4B

Qwen3.5-4B is a reasoning model from Alibaba. 9 benchmarks count toward its index, in 7 categories.

  • Open weights
  • Reasoning
  • Input: text
RankedEnough results for an index.
IndexAll categories, out of 100.44.3±7.4Median 51.0
CoverageIndex weight that has results.80%
SpeedOutput tokens per second.20/sMedian 67/s
First tokenSeconds to the first output token.102.5 sMedian 1.3 s
Input / 1MUS dollars per 1M input tokens.—Median $0.60
Output / 1MUS dollars per 1M output tokens.—Median $2.55
ContextMaximum tokens in one request.262KMedian 262K
EloLMArena rating and rank.—Median 1422

Run locally

4.66B parameters
16-bit
10.8 GB
8-bit
6.2 GB
4-bit
4.0 GB
Weights
Hugging Face (opens in a new tab) ↗

Memory need with a short context. The 4-bit copy fits in 8 GB. Compare local models

Capabilities

Index per category, out of 100.
Out of 100
AgenticMulti-step tasks with tools.1 counted63.9—
CodingCode writing and repair.None counted——
ReasoningLogic problems and puzzles.2 counted43.7146 / 227
MultimodalTasks with images and text.1 counted40.3—
KnowledgeFacts and expert knowledge.2 counted41.1189 / 268
MultilingualTasks in many languages.1 counted35.5—
InstructionTasks with strict prompt rules.1 counted37.6—
MathMath problems.1 counted41.4—

Results

9 counted
BenchmarkThe benchmark name.ResultThe score from the source.PlacePlace among results on this benchmark.
τ²-Bench Tool-Agent-User EvaluationAgenticAbove fitFar above this model's index.92.1%35 / 149
EuroEval ItalianMultilingual53.3% ±0.752 / 174
EuroEval SwedishMultilingual48.9% ±1.173 / 195
EuroEval PortugueseMultilingual48.7% ±0.674 / 182
EuroEval GermanMultilingual39.7% ±0.883 / 191
EuroEval PolishMultilingual44.2% ±0.894 / 209
EuroEval FrenchMultilingual51.5% ±0.582 / 178
EuroEval DutchMultilingual44.6% ±0.785 / 175
EuroEval SpanishMultilingual36.8% ±0.788 / 176
Artificial Analysis IFBenchInstruction52.0%78 / 146
Artificial Analysis GPQA DiamondKnowledge77.1%108 / 199
OTIS Mock AIME 2024-2025Math55.8% ±6.2115 / 200
Artificial Analysis Long Context ReasoningReasoning63.0%138 / 217
Artificial Analysis Humanity's Last ExamKnowledge9.9%145 / 222
Critical Physics TasksReasoning0.0%150 / 214
Artificial Analysis MMMU-ProMultimodal65.4%82 / 113

Sources

BenchLMCC BY-NC 4.0Epoch AICC BYEuroEvalMIT

History

5 releases
Index44.6 → 44.3
29 Oct 202612 Nov 2026

Questions

How good is Qwen3.5-4B?
Its index is 44.3 out of 100. That is place 183 of 283 ranked models.
What is Qwen3.5-4B best at?
Reasoning: 43.7 out of 100, place 146 of 227.
How fast is Qwen3.5-4B?
20 output tokens per second, 102.5 s to the first token.
What is the context window of Qwen3.5-4B?
262K tokens.
Is Qwen3.5-4B open source?
Yes. Its weights are published.

Badge

Qwen3.5-4B on Inference360: place 183 of 283, index 44.3
HTML
<a href="https://inference360.ai/models/qwen3-5-4b"><img src="https://inference360.ai/badges/qwen3-5-4b.svg" alt="Qwen3.5-4B on Inference360: place 183 of 283, index 44.3" height="20"></a>
Markdown
[![Qwen3.5-4B on Inference360: place 183 of 283, index 44.3](https://inference360.ai/badges/qwen3-5-4b.svg)](https://inference360.ai/models/qwen3-5-4b)

Compare with

DeepSeek V3.244.2Gemma 4 26B A4B44.5

More from Alibaba

All models
Qwen3.8 Max65.6Qwen3.8-Flash-Next64.3Qwen3.8 Max Preview64.2Qwen3.8 Max (0902)63.4Qwen3.8-Omni-Flash62.7