LifeVal evaluates personal assistants on everyday responsibilities, including caregiving, household administration, shopping, education, personal finances, and accessing government services. Scores are unweighted average rewards across 169 tasks, expressed as percentages, rather than a strict task-completion rate. The launch results cover five OpenAI models.
| Model / creator | Version | Reported score |
|---|---|---|
 GPT-6 AstraOpenAI · Proprietary | November 2026 | 80.6% |
| GPT-5.6OpenAI · Proprietary | November 2026 | 80% |
| GPT-5.5OpenAI · Proprietary | November 2026 | 77.4% |
| GPT-5.4OpenAI · Proprietary | November 2026 | 73% |
| GPT-5.2OpenAI · Proprietary | November 2026 | 49.3% |
aggregation: Average attempts within each task, then average all 169 task means equally. No willingness-to-pay weighting. Reward multiplied by 100; published display values rounded to one decimal place. · Launch-blog xhigh setting · tasks: 169
OpenAI ↗aggregation: Average attempts within each task, then average all 169 task means equally. No willingness-to-pay weighting. Reward multiplied by 100; published display values rounded to one decimal place. · Effort not specified in chart · tasks: 169
OpenAI ↗aggregation: Average attempts within each task, then average all 169 task means equally. No willingness-to-pay weighting. Reward multiplied by 100; published display values rounded to one decimal place. · Effort not specified in chart · tasks: 169
OpenAI ↗aggregation: Average attempts within each task, then average all 169 task means equally. No willingness-to-pay weighting. Reward multiplied by 100; published display values rounded to one decimal place. · Effort not specified in chart · tasks: 169
OpenAI ↗aggregation: Average attempts within each task, then average all 169 task means equally. No willingness-to-pay weighting. Reward multiplied by 100; published display values rounded to one decimal place. · Effort not specified in chart · tasks: 169
OpenAI ↗AppWorld evaluates agents completing everyday tasks across nine simulated apps through code and APIs.
AutomationBenchAutomationBench tests agents on realistic business workflows across applications and APIs, covering sales, marketing, operations, support, finance, and HR.
LiveBench: ResearchLiveBench 2.0 evaluates research reports on its October 25, 2026 release.
