The LLM benchmark aggregator.

Qwen3 235B A22B Instruct

Alibaba · 235B · A22B · open weights

Strongest in Safe-prompt compliance (#5 of 82), weakest in Jailbreak resistance (#217 of 272). Above par in 20 of 26 scopes. Among the models it meets almost everywhere, it finishes behind Claude Opus 5.5 and Claude Fable 5 and ahead of Claude Sonnet 4 and Gemini 3.5 Flash.

§ 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.

Core abilities55.7−11.7#38/2041/3

General intelligence58.2−11.9#38/2041/3

Coding52.7−16.4#53/1651/5

Code generation55.2−14.8#27/771/2

Agents55.8−12.3#58/2681/5

Tool use60.4−13.6#19/811/1

Professional51.8−9.8#63/1681/1

Medical53.8−13#52/1401/1

Legal51.5−15.6#66/1511/1

Finance51.2−11.7#81/1511/1

Safety54−7.5#71/3373/3

Safe-prompt compliance59.9−0.5#5/821/1

Secure code62.4−4.1#44/2741/1

Factual grounding51.7−18.9#47/1011/1

Fairness53.9−16.1#79/3002/2

Harm refusal54.5−7.1#80/3002/2

Toxicity avoidance53.1−5.9#150/2721/1

Jailbreak resistance40.4−24.5#217/2721/1

Knowledge50−13.5#85/1381/2

Academic knowledge50−13.9#78/1231/1

Human preference59.1−8.4#101/3421/1

Reasoning46.2−20.7#115/1783/4

Mathematics54.7−11#54/1401/2

Science45.9−17.6#88/1221/1

Reasoning39.8−29.3#112/1402/3

§ 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 235B A22B Instruct, left for the other.

DeepSeek V3won 24 · lost 1

Gemini 2.0 Flashwon 21 · lost 1

Mistral Large 3won 17 · lost 1

Gemini 1.5 Flashwon 16 · lost 1

GPT-4o Miniwon 21 · lost 2

Mistral Smallwon 19 · lost 2

Gemini 1.5 Prowon 17 · lost 2

Command R Pluswon 16 · lost 2

Llama 4 Maverick Instructwon 22 · lost 3

GPT-4.1 Nanowon 21 · lost 3

GPT-4.1 Miniwon 21 · lost 3

Command Awon 20 · lost 3

Mistral Largewon 16 · lost 3

Llama 4 Scout Instructwon 20 · lost 4

Gemini 2.5 Flash Litewon 16 · lost 4

GPT OSS 20Bwon 16 · lost 5

Grok 4.3won 17 · lost 6

GPT-4owon 16 · lost 6

Claude 3.5 Haikuwon 13 · lost 5

Grok 3won 15 · lost 6

Gemma 4 31Bwon 15 · lost 6

Ling 3.0 Flashwon 11 · lost 5

GPT-5.4 Nanowon 13 · lost 6

GLM-4.5won 13 · lost 6

Kimi K2won 17 · lost 8

GPT-5 Nanowon 16 · lost 8

Qwen3.7 Maxwon 15 · lost 8

Claude 3.5 Sonnetwon 13 · lost 7

Grok 4 Fastwon 11 · lost 6

Gemini 3.5 Flash Litewon 11 · lost 6

Claude Haiku 4.5won 16 · lost 9

GPT-4 Turbowon 10 · lost 6

GPT OSS 120Bwon 15 · lost 9

Mimo V2.5 Prowon 13 · lost 8

O4 Miniwon 16 · lost 10

GPT-5.4 Miniwon 11 · lost 7

GPT-4.1won 15 · lost 10

O3 Miniwon 13 · lost 9

Gemini 3.1 Flash Litewon 10 · lost 7

Qwen3.6 Pluswon 10 · lost 7

DeepSeek R1won 14 · lost 10

Kimi K2.6won 14 · lost 10

Gemini 3.6 Flashwon 13 · lost 10

Grok 4won 13 · lost 11

DeepSeek V3.2won 13 · lost 11

MiniMax M2.7won 10 · lost 9

GLM-4.6won 12 · lost 11

DeepSeek V4 Flashwon 12 · lost 11

Gemini 3.5 Flashwon 12 · lost 11

Claude Sonnet 4won 13 · lost 12

GLM-5.1won 10 · lost 10

Grok 4.1 Fastwon 11 · lost 11

MiniMax M3won 12 · lost 12

Gemini 2.5 Flashwon 13 · lost 13

GLM-5.2won 11 · lost 13

GPT-6 Lunawon 9 · lost 11

Kimi K2 Thinkingwon 8 · lost 11

Gemini 3.1 Prowon 8 · lost 11

Inklingwon 10 · lost 14

GPT-5.6 Lunawon 10 · lost 14

Claude Sonnet 4.6won 9 · lost 13

Kimi K2.5won 9 · lost 14

GLM-5.3 Flashwon 9 · lost 14

Claude 3.7 Sonnetwon 8 · lost 13

DeepSeek V4 Prowon 9 · lost 15

Claude Opus 4.6won 7 · lost 14

Grok 3 Miniwon 5 · lost 11

Qwen3.8 Maxwon 7 · lost 16

O1won 6 · lost 14

Claude Opus 4.5won 6 · lost 14

GLM-5won 5 · lost 12

Mimo V2.6 Prowon 5 · lost 12

GPT-5.6 Terrawon 7 · lost 17

GPT-5 Miniwon 7 · lost 18

Qwen3.8 27Bwon 5 · lost 13

Gemini 3 Prowon 6 · lost 16

Grok 4.5won 6 · lost 17

Muse Spark 1.1won 6 · lost 17

Gemini 3.8 Flashwon 6 · lost 17

Muse Spark 1.2won 4 · lost 12

Claude Opus 4won 6 · lost 18

Gemini 2.5 Prowon 6 · lost 19

Claude Sonnet 4.5won 6 · lost 19

GPT-6 Solwon 5 · lost 16

GPT-5.1won 5 · lost 17

Claude Sonnet 5won 5 · lost 19

O3won 5 · lost 20

GPT-5.2won 5 · lost 20

GLM-5.3won 3 · lost 14

GPT-5won 4 · lost 20

GPT-5.4won 4 · lost 20

GPT-5.6 Solwon 4 · lost 20

Claude Opus 4.7won 4 · lost 20

Grok 4.6won 3 · lost 20

DeepSeek V4.1 Flashwon 2 · lost 14

Gemini 3.7 Flashwon 2 · lost 15

Grok 4.7won 2 · lost 18

Claude Opus 4.8won 2 · lost 21

Claude Opus 5won 2 · lost 21

GPT-5.5won 2 · lost 22

GPT-6 Astrawon 1 · lost 15

Kimi K3won 1 · lost 16

Claude Fable 5.1won 0 · lost 17

Claude Fable 5won 0 · lost 17

Claude Opus 5.5won 0 · lost 20

§ 3 · Sources

Where the numbers come from

10 publications, 24 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-09-26

LMArena Text 1422

ARC-AGI-2 ↗1 measureread 2026-09-26

ARC-AGI-2 1.3%

LiveCodeBench ↗1 measureread 2026-09-26

LiveCodeBench 65.9%

BFCL ↗1 measureread 2026-09-26

BFCL v4 52.15%

Kagi LLM Benchmark ↗1 measureread 2026-09-26

Kagi LLM Benchmark 69.4%

SimpleBench ↗1 measureread 2026-09-26

SimpleBench 31%

Vals.ai ↗6 measuresread 2026-09-26

Vals · LegalBench 80.18%Vals · TaxEval 70.65%Vals · MedQA 90.62%Vals · GPQA Diamond 70.2%Vals · MMLU Pro 81.25%Vals · AIME 83.96%

HELM Safety ↗5 measuresread 2026-09-26

HELM Safety · HarmBench 79.6%HELM Safety · SimpleSafetyTests 100%HELM Safety · Anthropic Red Team 99.9%HELM Safety · BBQ 96.2%HELM Safety · XSTest 98.6%

Enkrypt AI Safety Leaderboard ↗6 measuresread 2026-09-26

Enkrypt · Jailbreak risk 20.9%Enkrypt · Harmful content risk 20%Enkrypt · CBRN risk 15.3%Enkrypt · Toxicity risk 4.1%Enkrypt · Bias risk 77.8%Enkrypt · Insecure code risk 8.4%

Vectara Hallucination Leaderboard ↗1 measureread 2026-09-26

Vectara · Factual consistency 90.7%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=qwen3-235b-a22b-instruct)](https://publicai.io/model-index/m/qwen3-235b-a22b-instruct)