Particle.news

OpenAI Releases GPT-6.2 Astra and Sol as Two-Tier Models for Workloads

The company says the pair cuts output token use, lowers Astra pricing, adds Ultrafast access for paid tiers, matches top third-party benchmarks, and carries self‑reported safety labels.

Overview

  • OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, Nov. 17, 2026, making both available in ChatGPT Work, Codex, and the API for Plus, Pro, Enterprise, and Edu customers.
  • Astra is positioned as the high-capability model for demanding professional work while Sol is a lower-cost model for routine coding and writing tasks and OpenAI says both use fewer output tokens.
  • OpenAI cut Astra’s per‑million-token price to $8 input and $40 output with cached input at $0.40, kept Sol at $2 input and $10 output with cached input at $0.10, and introduced Ultrafast token generation for select Pro, Enterprise, and Edu plans.
  • Independent evaluations show Astra at the frontier: a 174 score on the Epoch Capabilities Index tying Anthropic’s Claude Fable 5.2, a top AA‑Robotics score of 79 for real robot‑arm tasks, and Astra for Law leading Harvey’s LAB‑AA v1.1 with a 20.8% all‑pass rate while still failing most strict legal tasks.
  • OpenAI’s own Preparedness Framework rates Astra at 'High' for AI self‑improvement and Sol at 'Critical' for cybersecurity and 'High' for bio/chemical risks, and the company says those labels are guiding tighter access controls, partner API arrangements, and enterprise deployment choices.