Request a pilot
‹ PublicAI Index

The LLM benchmark aggregator.

Qwen3 4B Instruct

Alibaba · 4B

Strongest in Factual grounding (#21 of 101), weakest in Agents (#127 of 267). Above par in 2 of 4 scopes. Among the models it meets almost everywhere, it finishes behind GPT-4.1 and Gemini 2.5 Flash Lite and ahead of GPT-6 Astra and GPT-6 Sol.

§ 1 · Profile

What it is good at

Bars run from 50 — the average of the models each source lists — so right of the line is above par. The middle column is the gap to whoever leads that scope.

Safety53.4#90/336
Factual grounding60.8#21/101
Agents49.1#127/267
Tool use48.4#39/81

§ 2 · Head to head

What it beats, and what beats it

The same models turn up scope after scope. Counted once: where both were placed, who finished higher. Bars run right for Qwen3 4B Instruct, left for the other.

Ministral 3 3Bwon 3 · lost 0
Phi-4 Mini Instructwon 3 · lost 0
DeepSeek V3won 3 · lost 0
Llama 3.1 8B Instructwon 3 · lost 0
Ministral 3 8Bwon 3 · lost 0
NVIDIA Nemotron 3 Nano 30B A3Bwon 3 · lost 0
Llama 3.2 1B Instructwon 3 · lost 0
Ministral 3 14Bwon 3 · lost 0
DeepSeek R1won 3 · lost 0
GPT-4.1 Nanowon 3 · lost 0
Gemma 4 E4Bwon 3 · lost 0
Trinity Largewon 3 · lost 0
Gemini 3.1 Flash Litewon 3 · lost 0
Mistral Large 3won 3 · lost 0
Mercury 2won 3 · lost 0
Gemini 2.5 Prowon 3 · lost 0
Llama 3.2 3B Instructwon 3 · lost 0
Granite 3.2 8B Instructwon 3 · lost 0
Granite 3.1 8B Instructwon 3 · lost 0
Gemini 3.1 Prowon 3 · lost 0
Command R7Bwon 3 · lost 0
Qwen3.5 35B A3Bwon 3 · lost 0
Gemma 3 27B Itwon 4 · lost 0
Ministral 8Bwon 4 · lost 0
Llama 4 Scout Instructwon 4 · lost 0
Gemma 3 4B Itwon 4 · lost 0
Gemma 4 31Bwon 4 · lost 0
Gemma 3 12B Itwon 3 · lost 1
Llama 4 Maverick Instructwon 3 · lost 1
Nova Prowon 3 · lost 1
Qwen3 32Bwon 3 · lost 1
Qwen3 Next 80B A3B Thinkingwon 2 · lost 1
Palmyra Xwon 2 · lost 1
Qwen3.5 122B A10Bwon 2 · lost 1
Gemma 4 26B A4Bwon 2 · lost 1
GPT OSS 120Bwon 2 · lost 1
Kimi K2.5won 2 · lost 1
Kimi K2.6won 2 · lost 1
Claude Sonnet 4.6won 2 · lost 1
MiniMax M3won 2 · lost 1
DeepSeek V4 Prowon 2 · lost 1
GPT-5.6 Solwon 2 · lost 1
GPT-6 Solwon 2 · lost 1
GPT-6 Astrawon 2 · lost 1
Nova Microwon 2 · lost 2
Nova 2 Litewon 2 · lost 2
Qwen3 8Bwon 2 · lost 2
Phi 4won 2 · lost 2
Mistral Mediumwon 2 · lost 2
Command Awon 2 · lost 2
DeepSeek V3.2won 2 · lost 2
Gemini 2.5 Flashwon 2 · lost 2
Kimi K2won 2 · lost 2
GLM-4.6won 2 · lost 2
Grok 4.1 Fastwon 2 · lost 2
Gemini 3 Prowon 2 · lost 2
Claude Opus 4.5won 2 · lost 2
Claude Sonnet 4won 1 · lost 2
GPT-4.1 Miniwon 1 · lost 2
GPT-5.1won 1 · lost 2
Qwen3 30B A3B Instructwon 1 · lost 2
GPT-5won 1 · lost 2
Inklingwon 1 · lost 2
GPT-5.4won 1 · lost 2
GPT-5.5won 1 · lost 2
GLM-5.2won 1 · lost 2
Claude Opus 4.7won 1 · lost 2
Grok 4won 1 · lost 2
Qwen3 14Bwon 1 · lost 3
Mistral Smallwon 1 · lost 3
Mistral Largewon 1 · lost 3
GPT-5 Miniwon 1 · lost 3
Claude Haiku 4.5won 1 · lost 3
GPT-5 Nanowon 1 · lost 3
Qwen3 235B A22B Instructwon 1 · lost 3
O4 Miniwon 1 · lost 3
GPT-5.2won 1 · lost 3
Claude Sonnet 4.5won 1 · lost 3
GPT-5.4 Nanowon 0 · lost 3
GPT-5.4 Miniwon 0 · lost 3
O3won 0 · lost 3
GPT-5.6 Terrawon 0 · lost 3
GPT-5.6 Lunawon 0 · lost 3
Claude Sonnet 5won 0 · lost 3
Gemini 2.5 Flash Litewon 0 · lost 4
GPT-4.1won 0 · lost 4

§ 3 · Sources

Where the numbers come from

2 publications, 2 figures. Every one links to the page it was read from.

BFCL ↗1 measureread 2026-09-27
BFCL v4 35.68%
Vectara Hallucination Leaderboard ↗1 measureread 2026-09-27
Vectara · Factual consistency 94.3%

Badge

PublicAI Index badge for Qwen3 4B Instruct[![PublicAI Index](https://publicai.io/model-index/badge?model=qwen3-4b-instruct)](https://publicai.io/model-index/m/qwen3-4b-instruct)