Axion 1.0 - SEAL-0: leaderboard
Axion 1.0 author-run ReAct environment with web search, webpage visits, scholarly retrieval and Python. Best single run of three, or five for BioMysteryBench human-difficult; benchmark-approved website restrictions apply. Full HLE restricted websites and HLE-130 text-only are separate populations. Externally reported baseline scores are not pooled. SEAL-0.
Metric: Accuracy (%). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Axion 1.0-35B (Axion Run: harness=ReAct; selection=best%20of%203) | 62.2 | |
| 2 | Axion 1.0-9B (Axion Run: harness=ReAct; selection=best%20of%203) | 61.3 | |
| 3 | Axion 1.0-122B (Axion Run: harness=ReAct; selection=best%20of%203) | 58.6 | |
| 4 | Axion 1.0-4B (Axion Run: harness=ReAct; selection=best%20of%203) | 56.8 | |
| 5 | GPT-5.6 Sol (Axion Run: harness=ReAct; selection=best%20of%203) | 53.2 | |
| 6 | Gemini 3.1 Pro (Axion Run: harness=ReAct; selection=best%20of%203) | 53.2 | |
| 7 | Qwen3.5-122B-A10B (Axion Run: harness=ReAct; selection=best%20of%203) | 47.8 | |
| 8 | Qwen3.6-35B-A3B (Axion Run: harness=ReAct; selection=best%20of%203) | 44.1 |
Interactive version: theaggregate.ai/benchmark?slug=axion-1-0-seal-0 · How It Works · Data refreshed daily, snapshot 2026-11-05.