Atlas

Benchmarks

← All benchmarks

SnorkelWordle

General QA · 2026-06-08

SnorkelWordle is a language-game benchmark from Snorkel AI that tests models on 100 low-occurrence English Wordle words. Models have six attempts to identify each hidden five-letter word from feedback after each guess, and scores are reported as solve rate.

Top models (higher is better)

ModelScore
GPT-594.0
Grok 493.0
o392.9
O4 Mini91.9
Gemini 3 Pro Preview91.0
GPT-5 Mini91.0
o3-mini90.0
Grok 4 Fast88.0
Opus 485.6
Kimi K2 Thinking85.0
Sonnet 483.0
gpt-oss-120b81.6
GPT-5 Nano79.0
Gemini 2.5 Pro74.0
Grok 371.0
Claude 3.7 Sonnet68.0
gpt-oss-20b65.9
GPT-4.162.0
Gemini 2.5 Flash61.9
Kimi K2 Instruct54.0
Llama 3.3 70B Instruct10.2
Loading Atlas data…