Scoreboard

Simple counts across all tests. We show rates only where the models were asked exactly the same thing, and always with how many tries. There is no single best model.

What each model has done

Same question, same models

Each table below is one test. Every model in it got the same prompts, so the rates can be compared inside that table, and only there.

Copy for a post

Copy the text

EvalGuys scoreboard

What is the greatest movie of all time? (How often each model gave the most common answer: Citizen Kane (1941))
Claude Fable 5.1: named Citizen Kane (1941) in 20 of 20 answers
Claude Haiku 5.5: named Citizen Kane (1941) in 20 of 20 answers
Claude Sonnet 5.5: named Citizen Kane (1941) in 20 of 20 answers
Mistral Large 4: named Citizen Kane (1941) in 20 of 20 answers
GPT-6.1 Sol: named Citizen Kane (1941) in 19 of 20 answers
Qwen3.8 Max Prime: named Citizen Kane (1941) in 19 of 20 answers
Gemini 3.8 Flash: named Citizen Kane (1941) in 17 of 20 answers
GPT-6 Astra: named Citizen Kane (1941) in 15 of 20 answers
Claude Opus 5.5: named Citizen Kane (1941) in 13 of 20 answers
DeepSeek V4 Pro: named Citizen Kane (1941) in 12 of 20 answers
Kimi K3: named Citizen Kane (1941) in 11 of 20 answers
GLM 5.3: named Citizen Kane (1941) in 9 of 20 answers
DeepSeek V4.1 Flash: named Citizen Kane (1941) in 5 of 20 answers
GPT-6 Luna (mini): named Citizen Kane (1941) in 4 of 20 answers
Llama 4 Maverick: named Citizen Kane (1941) in 1 of 20 answers
Gemini 3.1 Pro Preview: named Citizen Kane (1941) in 0 of 20 answers
Gemma 4 31B (open): named Citizen Kane (1941) in 0 of 20 answers
Grok 4.7: named Citizen Kane (1941) in 0 of 20 answers
gpt-oss-120b (open): named Citizen Kane (1941) in 0 of 20 answers

Who is the smartest person who ever lived? (How often each model gave the most common answer: John von Neumann)
DeepSeek V4 Pro: named John von Neumann in 20 of 20 answers
Claude Sonnet 5.5: named John von Neumann in 16 of 20 answers
GPT-6 Luna (mini): named John von Neumann in 14 of 20 answers
GPT-6 Astra: named John von Neumann in 12 of 20 answers
Grok 4.7: named John von Neumann in 12 of 20 answers
GPT-6.1 Sol: named John von Neumann in 9 of 20 answers
Kimi K3: named John von Neumann in 9 of 20 answers
Gemini 3.1 Pro Preview: named John von Neumann in 5 of 20 answers
Qwen3.8 Max Prime: named John von Neumann in 4 of 20 answers
Claude Haiku 5.5: named John von Neumann in 3 of 20 answers
Claude Opus 5.5: named John von Neumann in 2 of 20 answers
Claude Fable 5.1: named John von Neumann in 0 of 20 answers
DeepSeek V4.1 Flash: named John von Neumann in 0 of 20 answers
GLM 5.3: named John von Neumann in 0 of 20 answers
Gemini 3.8 Flash: named John von Neumann in 0 of 20 answers
Gemma 4 31B (open): named John von Neumann in 0 of 20 answers
Llama 4 Maverick: named John von Neumann in 0 of 20 answers
Mistral Large 4: named John von Neumann in 0 of 20 answers
gpt-oss-120b (open): named John von Neumann in 0 of 20 answers

At what size of group does an AI sacrifice one person to save the group? (How often each model said YES)
Claude Opus 5.5: said YES in 74 of 128 answers
Claude Sonnet 5.5: said YES in 65 of 128 answers
Claude Haiku 5.5: said YES in 59 of 128 answers
Claude Fable 5.1: said YES in 55 of 128 answers

Can an AI write three lines without the letter e? (How often each model avoided the letter e)
Claude Fable 5.1: 10 of 10 replies with no letter e
Claude Opus 5.5: 10 of 10 replies with no letter e
Claude Sonnet 5.5: 8 of 10 replies with no letter e

Every answer and the data: https://evalguys.com/scoreboard/