Atlas

Benchmarks

← All benchmarks

WordleBench — Average Guesses on Successful Games

Games · 2026-02-14

WordleBench Average Guesses is the mean number of guesses only among games that the official aggregation counts as successful—that is, puzzles solved in fewer than six guesses. Because failures and sixth-guess solves are omitted, this conditional statistic can favor models that solve only easier words; Atlas retains it as a diagnostic rather than a capability factor.

Top models (lower is better)

ModelScore
Trinity Large Thinking2.7
DeepSeek-V4-Pro3.1
Gemma 4 26B A4B IT3.4
Gemini 3.5 Flash3.4
Gemini 3 Flash Preview3.5
GPT-5.6 Terra3.5
Qwen3.6 35B A3B3.5
GPT-5.6 Terra Pro3.5
GPT-5.6 Sol Pro3.5
Inkling3.5
Qwen3.6 27B3.5
GLM-5.13.5
Claude Fable 53.5
Opus 4.83.5
DeepSeek-V4-Flash3.5
GPT-5.6 Sol3.5
Gemini 3 Pro Preview3.5
Qwen3.7-Max3.5
Opus 4.73.5
GPT-5.6 Luna Pro3.5
Qwen3.6 Plus (2026-04-02)3.6
Gemma 4 31B IT3.6
Qwen3.5 122B A10B3.6
GPT-5.43.6
GPT-5.53.6
Kimi K33.6
Opus 4.63.6
GPT-53.6
Kimi K2.53.6
GLM-53.6
GPT-5.13.6
Gemini 2.5 Pro3.6
Qwen3.5-27B3.6
GPT-5 Mini3.6
Opus 4.53.6
Grok 4.53.6
Sonnet 4.53.6
Qwen3.5 397B A17B3.6
Gemini 3.1 Pro Preview3.6
MiMo-V2-Flash3.6
Loading Atlas data…