OpenAI LifeVal Benchmark: GPT-6 Astra Records 80.6%

·2026.11.04 11:30

Key point

OpenAI released the LifeVal benchmark, which measures the ability to perform everyday tasks.

1 / 4

Details

  • OpenAI released the LifeVal benchmark, which measures the ability to perform everyday tasks
  • GPT-6 Astra recorded the highest score at 80.6%, followed by GPT-5.6 at 80.0%
  • GPT-5.6 achieved similar performance at approximately half the API cost of GPT-6 Astra
  • LifeVal, consisting of 169 tasks, evaluates tasks centered on life in the United States

Performance and Cost Comparison

  • GPT-5.6 max recorded 82.6% performance at a cost of $9.65 with a 22.2-minute latency
  • GPT-6 Astra max recorded 82.5% performance at a cost of $20.04 with a 20.6-minute latency
  • Increasing reasoning intensity does not always increase the score; at lower intensities, newer models may score lower than older models

Limitations and Reproducibility

  • Reproducibility is low because the full task list, prompts, and dataset have not been released
  • There is a lack of comparison data with third-party models or human baselines

This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.

Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.