Atlas

Benchmarks

← All benchmarks

Eval Connections One-Shot Trap Bonus

Games · 2026-07-22

Trap-identification bonus as a percentage of the 40 available points on 20 canonical Eval Connections puzzles. Each correct first trap claim (or correct N/A on trap-free puzzles) contributes two points; invalid group submissions receive no bonus.

Top models (higher is better)

ModelScore
GPT-6 Astra85.0
Grok 4.680.0
Muse Spark 1.175.0
GPT-6 Sol75.0
Gemini 3.8 Flash70.0
Kimi K370.0
GPT-5.6 Sol70.0
Claude Opus 5.565.0
Gemini 3.1 Pro Preview65.0
GPT-5.6 Terra65.0
Claude Fable 565.0
Grok 4.765.0
Muse Spark 1.365.0
Gemini 3 Flash Preview60.0
GPT-5.6 Sol Pro60.0
Fugu Ultra v1.060.0
Claude Fable 5.160.0
GPT-5.6 Luna60.0
Grok 4.2055.0
Gemini 3.7 Flash50.0
DeepSeek V4 Pro 081350.0
Muse Spark 1.250.0
DeepSeek-V4-Flash-073150.0
Grok 4.550.0
GLM 5V Turbo50.0
DeepSeek V4.1 Flash50.0
GLM-5.3 Flash45.0
GPT-5.6 Terra Pro45.0
Qwen3.8 Flash45.0
MiMo-V2.6-Flash45.0
Qwen3.8 2.4T A95B45.0
Muse Glimmer 30B45.0
GPT-5.6 Luna Pro40.0
Sonnet 540.0
Gemini 3.5 Flash40.0
Inkling40.0
Claude Opus 535.0
Opus 4.535.0
Gemini 3.5 Flash-Lite35.0
Step 3.7 Flash35.0
Loading Atlas data…