TraderBench - Knowledge Retrieval: leaderboard
Metric: Knowledge Retrieval score (0-100) on 18 tasks drawn from BizFinBench and PRBench that require retrieving figures from SEC filings and market data servers (exact match plus LLM rubric), in the authors' two-agent setup (an A2A Candidate Agent with the same six financial MCP servers for every model; GPT-5.2 judge at temperature 0 for rubric-scored sections); higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.1 Fast | 61.6 | #208 |
| 2 | Gemini 3 Pro | 52.4 | #77 |
| 3 | GPT-5.2 | 50.6 | #105 |
| 4 | GPT-OSS-20B | 42.4 | #499 |
| 5 | GLM-4.7 Flash | 38.9 | #496 |
| 6 | Kimi K2.5 | 37.1 | #139 |
| 7 | GPT-OSS-120B | 32.5 | #330 |
| 8 | GPT-4o | 32.3 | #333 |
| 9 | Qwen 3 32B (Non-reasoning) | 14.4 | #424 (Qwen 3 32B) |
| 10 | Qwen 3 8B | 10.5 | #667 |
| 11 | Qwen 3 30B A3B | 9.4 | #488 |
| 12 | Gemma 3 27B | 9.3 | #596 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=traderbench-knowledge-retrieval · How It Works · Data refreshed daily, snapshot 2026-10-11.