Atlas

Benchmarks

← All benchmarks

PACT Negotiation CMS

Agents · 2025-07-29

PACT Composite Model Score combines opponent-balanced performance with captured economic surplus using the benchmark's fixed alpha blend. Higher CMS indicates stronger negotiating performance in the public leaderboard.

Top models (higher is better)

ModelScore
Claude Fable 569.0
GPT-5.561.0
Opus 4.856.0
DeepSeek-V4-Pro54.0
GLM-5.254.0
Gemini 3.1 Pro Preview54.0
Mistral Medium 3.552.0
Gemma 4 31B IT50.0
Sonnet 4.650.0
Kimi K2.649.0
GLM-5.149.0
Hy3 preview-Base49.0
gpt-oss-120b48.0
Gemini 3.1 Flash-Lite Preview47.0
Haiku 4.546.0
Doubao Seed 2.0 Pro45.0
Trinity Large Thinking44.0
MiMo-V2.5-Pro44.0
Qwen3.7-Max41.0
Grok 4.340.0
Step 3.7 Flash37.0
DeepSeek-V4-Flash37.0
ERNIE 5.132.0
Nova Pro32.0
Mistral Large 3 675B Instruct 251228.0
Llama 4 Maverick26.0
Loading Atlas data…