Benchmarks · agentic
RSI-Exam 0.1 (aiming-lab)
An agent spends a long, budgeted run improving an inherited, working-but-weak executable research method or harness on a visible split, and the single artifact it submits is re-run once on a sealed hidden split across 88 executable research tasks in six domains. Maintained by RSI-Exam Team. Score: Mean frozen normalised hidden-set score (0–1) over the 88 active tasks (Full board).
13 models · 13 results · data as of 2026-10-29 · rsi-exam.ai ↗
Leaderboard
GPT-6 Astra Codex max
All results
13 rows| # | Model | Effort | Score | Harness |
|---|---|---|---|---|
| 1 | GPT-6 Astra Codex | max | 51.3% | codex |
| 2 | Fable 5.1 | max | 48.1% | claude code |
| 3 | Opus 5 | max | 46.4% | claude code |
| 4 | GPT-5.6 Sol Codex | max | 43.3% | codex |
| 5 | GLM 5.3 | max | 40.3% | claude code |
| 6 | Qwen 3.8 Max | xhigh | 39.2% | qwen coder |
| 7 | Muse Spark 1.3 | max | 39.1% | musecode |
| 8 | Kimi K3 | max | 38.2% | kimi cli |
| 9 | Grok 4.6 | xhigh | 36.7% | grok |
| 10 | Gemini 3.8 Flash | high | 34.1% | antigravity |
| 11 | GPT-5.5 Codex | xhigh | 33.1% | codex |
| 12 | DeepSeek V4 Pro | max | 32.3% | claude code |
| 13 | Gemini 3.7 Flash | high | 30.9% | antigravity |