OpenAI LifeVal: leaderboard
OpenAI's LifeVal (2026-11-02; paper arXiv 2611.06687): 169 personalized consumer tasks drawn from everyday life (managing finances, moving apartments, job search, finding childcare) that an agent completes in reinforcement-learning environments with an offline-web browser tool. Unlike general web-agent suites, credit depends on using the task writer's own ChatGPT conversations, memories, files and context. A GPT-5.6 Sol (xhigh) grader awards rubric credit, environment hacking scores zero, attempts are averaged per task and tasks weigh equally. Average reward (%), higher is better; one row per model and reasoning effort.
Metric: Average 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 | #5 (GPT-6 Astra) |
| 3 | GPT-6 Astra (High) | 80.72 | #5 (GPT-6 Astra) |
| 4 | GPT-6 Astra (xHigh) | 80.58 | #5 (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 | #5 (GPT-6 Astra) |
| 8 | GPT-5.5 (xHigh) | 77.45 | #34 (GPT-5.5) |
| 9 | GPT-5.5 (High) | 75.61 | #34 (GPT-5.5) |
| 10 | GPT-5.6 Sol (Medium) | 75 | #22 (GPT-5.6 Sol) |
| 11 | GPT-5.4 (xHigh) | 73.05 | #85 (GPT-5.4) |
| 12 | GPT-5.4 (High) | 72 | #85 (GPT-5.4) |
| 13 | GPT-5.5 (Medium) | 71.31 | #34 (GPT-5.5) |
| 14 | GPT-5.4 (Medium) | 70.14 | #85 (GPT-5.4) |
| 15 | GPT-6 Astra (Low) | 66.59 | #5 (GPT-6 Astra) |
No result here: #3 Claude Opus 5.5, #6 GPT-6.1 Sol, #9 Gemini 4 Argon.
Interactive version: theaggregate.ai/benchmark?slug=openai-lifeval · How It Works · Data refreshed daily, snapshot 2026-11-07.