Share your thoughts, 1 month free Claude Pro on usSee more

LLM Chat Evaluation on ArenaHard

49.2Accuracy

Official Instruct Model

Updated 1d ago

Evaluation Results

Method	Links
Official Instruct Model 2026.05		49.2
GLM-4.7 (Scaffold-only) 2026.05		17.32
GPT-5.4 (High) 2026.05		14.34
GLM-4.7 & ANDES (Ours) 2026.05		14.11
Opus-4.7 (xHigh) 2026.05		10.21
Opus-4.6 (1M) 2026.05		9
Opus-4.6 2026.05		7.78
GPT-5.2 2026.05		6.49
Gemini-3.1-Pro 2026.05		6.42
MiniMax-M2.5 2026.05		2.18
Base Model (SmolLM3-3B) 2026.05		0.42
Kimi-K2-Thinking 2026.05		0.42
MiniMax-M2.1 2026.05		0.42
GPT-5.1-Codex-Max 2026.05		0.42
GLM-4.7 (OpenCode) 2026.05		0.42
Sonnet-4.5 2026.05		0.24
Qwen3-Max 2026.05		0.21