LifeVal - Mean Task Reward: leaderboard
LifeVal November3 protocol:169 personalized consumer tasks with RLEs, browser access to an offline but non-frozen web cache, conversation history, memories and files;8-hour rollout limit. GPT5.6Sol xhigh primary grader independently scores writer rubric items using final answer, artifacts and backend-state changes; credit only for completed actions. Internal-RLEAPI hacking receives zero. Native reward averages attempts within each task, then the169task means equally, without willingness-to-pay weights. Failed/incomplete actions receive no completion credit; no blanket zero or retry rule beyond published grading is inferred. Exact model and reasoning effort retained; time, tokens, APIcost and human grading-calibration scores are diagnostics, not capability rows.
Metric: Mean task reward (%). Source: openai.com. Saturation forecast: Around February 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.6 Sol (Max) | 82.58 | #22 (GPT-5.6 Sol) |
| 2 | GPT-6 Astra (Max) | 82.48 | #6 (GPT-6 Astra) |
| 3 | GPT-6 Astra (High) | 80.72 | #6 (GPT-6 Astra) |
| 4 | GPT-6 Astra (xHigh) | 80.58 | #6 (GPT-6 Astra) |
| 5 | GPT-5.6 Sol (xHigh) | 79.98 | #22 (GPT-5.6 Sol) |
| 6 | GPT-5.6 Sol (High) | 78.81 | #22 (GPT-5.6 Sol) |
| 7 | GPT-6 Astra (Medium) | 77.61 | #6 (GPT-6 Astra) |
| 8 | GPT-5.5 (xHigh) | 77.45 | #33 (GPT-5.5) |
| 9 | GPT-5.5 (High) | 75.61 | #33 (GPT-5.5) |
| 10 | GPT-5.6 Sol (Medium) | 75 | #22 (GPT-5.6 Sol) |
| 11 | GPT-5.4 (xHigh) | 73.05 | #83 (GPT-5.4) |
| 12 | GPT-5.4 (High) | 72 | #83 (GPT-5.4) |
| 13 | GPT-5.5 (Medium) | 71.31 | #33 (GPT-5.5) |
| 14 | GPT-5.4 (Medium) | 70.14 | #83 (GPT-5.4) |
| 15 | GPT-6 Astra (Low) | 66.59 | #6 (GPT-6 Astra) |
No result here: #3 Claude Opus 5.5, #7 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=lifeval-mean-task-reward · How It Works · Data refreshed daily, snapshot 2026-11-05.