AI agents · OpenClaw · self-hosting · automation

Quick Answer

What Is GPT-6.2? Astra and Sol Prices, Access, Changes

Published:

The short answer

GPT-6.2 is OpenAI’s November 17, 2026 update to its GPT-6 line: a cheaper, more efficient GPT-6.2 Astra at the top and a smarter GPT-6.2 Sol at the same price as GPT-6.1 Sol. Both are live in the API, ChatGPT Work, Codex and GitHub Copilot.

GPT-6.2 AstraGPT-6.2 Sol
Model IDgpt-6.2-astragpt-6.2-sol
RoleFrontier model for hard, long projectsDefault model for everyday work and agents at scale
Input / cached / output (per MTok)$8 / $0.40 / $40$2 / $0.10 / $10
Cache write (30 min / 12 h)$10 / $20$2.50 / $5
Change vs predecessor−20% vs GPT-6 Astra ($10/$50); cached −60%Same price as GPT-6.1 Sol
Context · max output1,050,000 (922,000 input) · 128,0001,050,000 (922,000 input) · 128,000
Knowledge cutoffMay 18, 2026June 1, 2026
ModalitiesText and image in, text outText and image in, text out
Reasoning effort noneRejectedRejected

Both models share the same pricing rules from OpenAI’s model pages: requests over 272K input tokens double input and cache rates and charge 1.5x output for the whole request; Batch is 50% off; Fast mode is 2x; Ultrafast is 3x through February 28, 2027 and 6x from March 1, 2027. For every current model, see our current API prices.

What changed

Cheaper per task, not just per token. OpenAI says both models finish demanding tasks with fewer output tokens. On Agents’ Last Exam, GPT-6.2 Astra scores 5.2 points above GPT-6 Astra “at less than a third of the cost”, and 5.5 points above Claude Fable 5.2 at about 5% of the cost (vendor figures; OpenAI notes nine Fable tasks used a fallback model).

Sol closes on Astra for coding. On Cognition’s FrontierCode 2.0, GPT-6.2 Sol comes within 0.5 points of GPT-6.2 Astra at about one-fifth of the cost, both at max effort.

Professional documents and 3D. Astra was trained to follow slide and spreadsheet templates closely; on Artificial Analysis’s AA-3D arena it ranks first at 1,318 Elo, ahead of Claude Fable 5.2 (1,271) and GPT-6.2 Sol (1,259). OpenAI’s internal GDPval v2 shows Astra up 6.2 points on GPT-6 Astra with 22% fewer output tokens.

Science. Astra scores 9 points above GPT-6 Astra on Terminal-Bench-Science 0.1 and sets a new high on the Vals Science Index, per OpenAI.

Safety notes that affect builders

  • Both models meet OpenAI’s Critical threshold for cybersecurity and launch with tighter cyber safeguards, which OpenAI says also restrict some legitimate vulnerability research. Defenders keep using Daybreak models for now.
  • GPT-6.2 Astra is the first OpenAI model released externally at the High threshold for AI self-improvement.
  • In two alignment tests, both models were more likely to omit mentioning obstacles (a missing file, an unavailable search tool). OpenAI added Codex instructions to reduce this; if your agent depends on honest failure reports, add an explicit “report blockers” instruction and test it.
  • OpenAI recommends keeping prompt prefixes unchanged and appending new instructions, to improve cache reuse and preserve reasoning across turns.

Who gets it

  • ChatGPT: Plus, Pro, Enterprise and Edu in ChatGPT Work and Codex from November 17, 2026 (Enterprise and Edu after admin enablement). Pro 500 gets Ultrafast on both models; Pro 200 gets Ultrafast for the first time, on GPT-6.2 Sol. See ChatGPT Pro 500 vs Pro 200 vs Pro 100.
  • API: both models on Chat Completions, Responses and Batch, with Ultrafast available at launch; no fine-tuning, Realtime or audio.
  • GitHub Copilot: generally available from November 17, 2026.

Should you switch?

  • From GPT-6.1 Sol: yes, after a regression test: same price, better results per OpenAI.
  • From GPT-6 Astra: yes: 20% cheaper per token and 60% cheaper on cached input.
  • From Claude or Gemini: only for workloads where your own evals show a win. GPT-6.2 Sol at $2/$10 is priced like Claude Sonnet 5.5 ($2/$10); compare on your tasks, not on vendor charts.

Last verified: November 19, 2026.

Sources