Atlas

Benchmarks

← All benchmarks

Antidote: Everyday Edition

Safety · 2026-05-21

Elo rating on real-world everyday prompts graded by experts.

Top models (higher is better)

ModelScore
Claude Fable 51099
Gemini 3.1 Pro Preview1091
Gemini 3.8 Flash1087
Gemini 3.6 Flash1083
Gemini 3.5 Flash1079
Kimi K31077
Gemini 3.7 Flash1075
GLM-5.31067
Claude Fable 5.11065
Claude Opus 51055
Qwen3.7-Max (source unspecified)1051
Kimi K2.61045
Opus 4.61042
Opus 4.71042
GLM-5.21040
Opus 4.81030
Muse Spark 1.21028
GPT-5.6 Sol1027
Hy31022
DeepSeek V4 Flash Vision Exp1015
Muse Spark 1.11014
GPT-5.51013
Sonnet 4.61012
DeepSeek-V4-Pro1007
Kimi K2.51006
Grok 4.201001
Qwen3.5 Plus (2026-02-15)999
Gemini 3.5 Flash-Lite991
Inkling991
Grok 4.5973
DeepSeek-V4-Flash966
Grok 4.6966
DeepSeek-V3.2954
Grok 4.3950
ERNIE 5.1947
Mistral Large 3 675B Instruct 2512947
Haiku 4.5932
Muse Glimmer 30B931
GPT-5.4 Mini929
Gemma 3 12B IT893
Loading Atlas data…