PACT (Lechmazur): leaderboard
Pairwise Auction Conversation Testbed for multi-round buyer-seller bargaining. LLMs negotiate over 20 rounds with hidden private values, scored by Composite Model Score from head-to-head surplus capture.
Metric: PACT Bilateral Rating. Source: github.com. Saturation forecast: Around July 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 (High) | 1607 | #31 (GPT-5.5) |
| 2 | Claude Fable 5 (High) | 1603 | #16 (Claude Fable 5) |
| 3 | DeepSeek V4 Pro | 1570 | #102 |
| 4 | Claude Opus 4.8 (High) | 1562 | #37 (Claude Opus 4.8) |
| 5 | Kimi K2.6 | 1560 | #104 |
| 6 | Gemini 3.1 Pro (Preview) | 1557 | #58 |
| 7 | Gemma 4 31B (Reasoning) | 1557 | #200 (Gemma 4 31B) |
| 8 | GLM-5.1 | 1554 | #97 |
| 9 | Claude Sonnet 4.6 (High) | 1546 | #90 (Claude Sonnet 4.6) |
| 10 | Mistral Medium 3.5 (High) | 1538 | #353 (Mistral Medium 3.5) |
| 11 | Trinity-Large-Thinking | 1526 | #356 |
| 12 | Hy3-preview | 1520 | #157 |
| 13 | Qwen 3.7 Max | 1511 | #75 |
| 14 | Claude Haiku 4.5 | 1510 | #280 |
| 15 | Gemini 3.1 Flash Lite (Preview) | 1509 | #194 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=pact-lechmazur · How It Works · Data refreshed daily, snapshot 2026-10-20.