GPT-6.2 Astra & GPT-6.2 Sol: OpenAI Tops the Pareto Frontier

Matt Weaver, Head of Applied AI Architects at OpenAI, explains how GPT-6.2 Astra and Sol deliver stronger enterprise outcomes while reducing token usage

GPT-6.2 Astra & GPT-6.2 Sol: OpenAI Tops the Pareto Frontier

TL;DR

  • OpenAI introduces GPT-6.2 Astra (for complex tasks) and GPT-6.2 Sol (for everyday tasks).
  • These models aim to deliver stronger enterprise outcomes with reduced token usage.
  • OpenAI ranks second in the Artificial Analysis Intelligence Index, behind Anthropic's Claude Fable 5.2.
  • OpenAI models now occupy a significant portion of the Cost per Task Pareto frontier.
  • GPT-6.2 Astra shows improvements on benchmarks like Agents' Last Exam and OpenAI's internal GDPval v2.
  • GPT-6.2 Sol demonstrates cost-effectiveness, outperforming GPT-6 Astra on the LifeVal evaluation at a fraction of the cost.
  • API input and output token prices for GPT-6.2 Astra are 20% lower than GPT-6 Astra.
  • GPT-6.2 Astra meets High thresholds for cybersecurity and biological/chemical capabilities, and for AI Self-Improvement.
  • GPT-6.2 Sol meets High thresholds for cybersecurity and biological/chemical capabilities but falls below the AI Self-Improvement threshold.