The LLM benchmark aggregator.

Claude 3.5 Haiku

Anthropic

Strongest in Jailbreak resistance (#5 of 272), weakest in Secure code (#217 of 274). Above par in 5 of 20 scopes. Among the models it meets almost everywhere, it finishes behind Claude Fable 5 and Claude Fable 5.1 and ahead of Gemini 1.5 Pro and Grok Build 0.1.

§ 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.

Safety58.8−2.7#13/3371/3

Jailbreak resistance63.8−1.1#5/2721/1

Harm refusal60.1−1.5#8/3001/2

Fairness70leads#10/3001/2

Toxicity avoidance58.1−0.9#27/2721/1

Secure code40.7−25.8#217/2741/1

Coding46.5−22.6#105/1651/5

Agentic coding45.7−23#102/1571/4

Knowledge32.2−31.3#133/1381/2

Academic knowledge28.6−35.3#117/1231/1

Professional43.3−18.3#140/1681/1

Legal44−23.1#110/1511/1

Finance39.6−23.3#131/1511/1

Reasoning33.8−33.1#175/1781/4

Science26.1−37.4#116/1221/1

Mathematics33.3−32.4#135/1401/2

Human preference49.8−17.7#194/3421/1

Agents44.1−24#199/2681/5

Knowledge work41.1−32.5#127/1781/2

§ 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 Claude 3.5 Haiku, left for the other.

Jamba 1.6 Miniwon 14 · lost 0

Command R Pluswon 13 · lost 1

Jamba 1.6 Largewon 12 · lost 2

GPT-4.1 Nanowon 16 · lost 4

GPT-4o Miniwon 15 · lost 4

Gemma 4 E4Bwon 9 · lost 3

Gemma 3 27B Itwon 9 · lost 3

GPT-4 Turbowon 8 · lost 4

Llama 4 Scout Instructwon 12 · lost 6

Gemini 1.5 Flashwon 10 · lost 6

Nemotron 3 Ultra 550B A55Bwon 7 · lost 5

Grok Build 0.1won 7 · lost 5

Gemini 1.5 Prowon 9 · lost 7

GPT OSS 20Bwon 9 · lost 9

Llama 4 Maverick Instructwon 10 · lost 10

DeepSeek R1won 9 · lost 10

Ling 3.0 Flashwon 8 · lost 9

Kimi K2 Thinkingwon 7 · lost 9

Mistral Smallwon 7 · lost 9

Llama 3.3 Instruct Turbo (70B)won 6 · lost 8

Command Awon 8 · lost 11

Gemma 4 12Bwon 5 · lost 7

Kimi K2.7 Codewon 5 · lost 7

O1 Miniwon 5 · lost 7

Gemini 3.1 Flash Litewon 6 · lost 9

GPT OSS 120Bwon 8 · lost 12

DeepSeek V3won 8 · lost 12

Kimi K2.5won 8 · lost 12

GPT-4owon 7 · lost 11

Gemini 2.0 Flashwon 7 · lost 11

Mistral Largewon 6 · lost 10

Grok 4.1 Fastwon 7 · lost 12

Mistral Medium 3.5won 5 · lost 9

O3 Miniwon 7 · lost 13

GPT-5 Miniwon 7 · lost 13

Claude Haiku 4.5won 7 · lost 13

GLM-4.6won 7 · lost 13

MiniMax M3won 7 · lost 13

Grok 4.3won 7 · lost 13

Gemini 2.5 Prowon 7 · lost 13

Qwen3.7 Maxwon 7 · lost 13

Mistral Large 3won 5 · lost 10

Mimo V2.6 Flashwon 5 · lost 10

Gemma 4 31Bwon 6 · lost 12

GLM-4.5won 5 · lost 11

Grok 3won 5 · lost 11

MiniMax M2.7won 6 · lost 14

GPT-5 Nanowon 5 · lost 12

GPT-6 Lunawon 5 · lost 12

GPT-6 Solwon 5 · lost 12

Mimo V2.6 Prowon 5 · lost 12

Qwen3 235B A22B Instructwon 5 · lost 13

Claude Opus 4won 5 · lost 13

GPT-5.4 Nanowon 4 · lost 11

GPT-5.4 Miniwon 4 · lost 11

Claude 3.7 Sonnetwon 5 · lost 14

O4 Miniwon 5 · lost 14

Gemini 2.5 Flashwon 5 · lost 14

Grok 4won 5 · lost 14

GPT-4.1won 5 · lost 14

Kimi K2won 5 · lost 14

O3won 5 · lost 14

GPT-5.2won 5 · lost 14

GPT-4.1 Miniwon 5 · lost 15

DeepSeek V3.2won 5 · lost 15

DeepSeek V4 Flashwon 5 · lost 15

Inklingwon 5 · lost 15

GPT-5.6 Lunawon 5 · lost 15

GPT-5.1won 5 · lost 15

Claude Sonnet 4.5won 5 · lost 15

Kimi K2.6won 5 · lost 15

DeepSeek V4 Prowon 5 · lost 15

GLM-5.1won 5 · lost 15

Grok 4.5won 5 · lost 15

GPT-5.6 Terrawon 5 · lost 15

Mimo V2.5 Prowon 5 · lost 15

GLM-5.3 Flashwon 5 · lost 15

GPT-5.4won 5 · lost 15

GLM-5.2won 5 · lost 15

Gemini 3.5 Flashwon 5 · lost 15

Qwen3.8 Maxwon 5 · lost 15

GPT-5.5won 5 · lost 15

Gemini 3.6 Flashwon 5 · lost 15

GPT-5.6 Solwon 5 · lost 15

Gemini 3.8 Flashwon 5 · lost 15

Grok 4 Fastwon 3 · lost 10

Gemini 3 Flashwon 3 · lost 10

O1won 4 · lost 14

Gemini 2.5 Flash Litewon 3 · lost 11

Gemini 3.5 Flash Litewon 3 · lost 11

Claude Opus 4.5won 3 · lost 11

Claude Sonnet 4won 4 · lost 16

GPT-5won 4 · lost 16

Grok 4.6won 4 · lost 16

Claude Opus 4.8won 4 · lost 16

Claude Opus 5won 4 · lost 16

Muse Spark 1.1won 4 · lost 16

Claude Opus 4.7won 4 · lost 16

Gemini 3 Prowon 3 · lost 13

Claude Opus 4.6won 3 · lost 13

Grok 4.7won 3 · lost 14

Claude Opus 5.5won 3 · lost 14

Qwen3 Maxwon 2 · lost 10

Claude Sonnet 4.6won 3 · lost 15

Qwen3.6 Pluswon 2 · lost 12

Gemini 3.1 Prowon 2 · lost 13

Claude Sonnet 5won 2 · lost 18

GPT-6 Astrawon 1 · lost 11

Qwen3.5 Flashwon 1 · lost 12

GLM-5won 1 · lost 12

Mimo V2.5won 1 · lost 13

Claude 3.5 Sonnetwon 1 · lost 16

MiniMax M2.1won 0 · lost 12

MiniMax M2.5won 0 · lost 12

DeepSeek V4.1 Flashwon 0 · lost 13

Muse Spark 1.2won 0 · lost 13

Inkling Smallwon 0 · lost 14

Qwen3.8 27Bwon 0 · lost 14

GLM-4.7won 0 · lost 14

GLM-5.3won 0 · lost 14

Gemini 3.7 Flashwon 0 · lost 14

Kimi K3won 0 · lost 14

Muse Sparkwon 0 · lost 14

Claude Fable 5.1won 0 · lost 14

Claude Fable 5won 0 · lost 14

§ 3 · Sources

Where the numbers come from

5 publications, 15 figures. Every one links to the page it was read from.

LMArena Text ↗1 measureread 2026-09-26

LMArena Text 1325

GDPval-AA ↗1 measureread 2026-09-26

GDPval-AA 198

Aider polyglot ↗1 measureread 2026-09-26

Aider polyglot 28%

Vals.ai ↗6 measuresread 2026-09-26

Vals · LegalBench 70.33%Vals · CorpFin 50.82%Vals · TaxEval 57.36%Vals · GPQA Diamond 37.88%Vals · MMLU Pro 64.12%Vals · AIME 3.33%

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

Enkrypt · Jailbreak risk 1.1%Enkrypt · Harmful content risk 5.6%Enkrypt · CBRN risk 3%Enkrypt · Toxicity risk 0.6%Enkrypt · Bias risk 43.4%Enkrypt · Insecure code risk 52.4%

Badge

[![PublicAI Index](https://publicai.io/model-index/badge?model=claude-3-5-haiku)](https://publicai.io/model-index/m/claude-3-5-haiku)