C2A Bench - Redwood Inference: leaderboard
Observe.AI C2A Bench (observeai/c2a-bench) Redwood Inference domain: builder models inside coding-agent harnesses build and maintain a customer-support agent across 6 evolving documentation stages (27 to 80 docs, 290 graded tickets) with runtime agent fixed to openai/gpt-5.6-sol. Mean ticket pass rate (0-100%); higher is better.
Metric: Ticket Pass Rate (%). Source: huggingface.co. Saturation forecast: Not forecast. 7 models tracked. Updated 21 Nov 2026.
Top models
| # | Model | Setting | Score | Overall rank |
|---|---|---|---|---|
| 1 | GPT-6.1 Sol | High | 80.07 | #15 |
| 2 | DeepSeek V4.1 Flash (1017) | High | 77.46 | #47 |
| 3 | Claude Opus 5.5 | High | 76.78 | #9 |
| 4 | GLM-5.3 | High | 75.16 | #72 |
| 5 | Claude Sonnet 5.5 | High | 71.35 | #24 |
| 6 | GPT-6 Astra | High | 67.2 | #10 |
| 7 | Gemini 4 Carbon | High | 63.59 | #7 |
No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.
Interactive version: theaggregate.ai/benchmark?slug=c2a-bench-redwood-inference · How It Works · Data refreshed daily, snapshot 2026-11-21.