FrontierML ($200 Wallet): leaderboard
Induction Labs evaluates coding agents on 21 end-to-end machine learning tasks spanning language, vision, audio and games. Each task allows 12 hours and a $200 wallet for training compute and purchased model calls; the coding agent's own API costs are additional. Agents choose architectures, obtain data, train models and submit an inference server. Held-out evaluation and a separate compliance review produce a continuous score. This board is the published mean across all 21 tasks, on a 0-100 scale, at the fixed $200 budget. The evaluated harness and reasoning effort remain part of each model identity.
Metric: Mean Task Score (%). Source: www.frontierml.org. Saturation forecast: Around January 2028. 14 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5.5 (xHigh) | 58.85 | #3 (Claude Opus 5.5) |
| 2 | Claude Sonnet 5.5 (xHigh) | 50.45 | #10 (Claude Sonnet 5.5) |
| 3 | GPT-6 Astra (xHigh) | 44.53 | #5 (GPT-6 Astra) |
| 4 | Claude Fable 5.1 (xHigh) | 41.91 | #9 (Claude Fable 5.1) |
| 5 | GPT-6.1 Sol (xHigh) | 38.67 | #8 (GPT-6.1 Sol) |
| 6 | Claude Haiku 5.5 (Max) | 34.93 | #35 (Claude Haiku 5.5) |
| 7 | Kimi K3 (Max) | 20.33 | #30 (Kimi K3) |
| 8 | GLM-5.3 (Max) | 19.63 | #40 (GLM-5.3) |
| 9 | Qwen 3.8 Max (xHigh) | 17.47 | #31 (Qwen 3.8 Max) |
| 10 | GPT-6 Luna (xHigh) | 13.12 | #52 (GPT-6 Luna) |
No result here: #7 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=frontierml-200-wallet · How It Works · Data refreshed daily, snapshot 2026-10-16.