BAZAAR (Lechmazur): leaderboard
Lech Mazur's BAZAAR: LLMs act as buyers and sellers in a sealed-quote double-auction market, quoting from private values and past rounds without seeing the order book. Ratings are TrueSkill μ from 200 shuffled passes over about 2,000 games, derived from conditional surplus alpha versus a truthful baseline; higher is better. Scripted bot baselines are omitted.
Metric: TrueSkill μ. Source: github.com. Saturation forecast: Estimated already saturated. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | O3 (Medium) | 7.37 | #132 (O3) |
| 2 | Gemini 2.5 Pro | 7.2 | #159 |
| 3 | Gemini 2.5 Flash | 7.16 | #255 |
| 4 | Claude Sonnet 4 (Non-reasoning) | 7.09 | #207 (Claude Sonnet 4) |
| 5 | Grok 4 | 6.91 | #189 |
| 6 | Claude Opus 4 (Thinking 16K) | 6.86 | #170 (Claude Opus 4) |
| 7 | O4 Mini (Medium) | 6.77 | #185 (O4 Mini) |
| 8 | DeepSeek R1 0528 | 6.55 | #232 |
| 9 | Claude Opus 4 (Non-reasoning) | 6.53 | #170 (Claude Opus 4) |
| 10 | Qwen 3 235B A22B | 6.52 | #323 |
| 11 | GPT-4o (Mar 2025) | 5.89 | #338 |
| 12 | Llama 4 Maverick | 5.51 | #486 |
| 13 | DeepSeek V3 (0324) | 5.22 | #349 |
| 14 | Nova Pro | 5.22 | #594 |
| 15 | Claude 3.5 Haiku | 5.03 | #571 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=bazaar-lechmazur · How It Works · Data refreshed daily, snapshot 2026-11-05.