ModelBrief · GPT-6.2 Astra · Topic Page
Last updated Nov 20, 2026

GPT-6.2 Astra Access and API Guide: IDs, Plans, System Card

ModelBrief · Published 2026-11-18 · Fact-check 2026-11-18 · Reference: OpenAI model page, OpenAI launch post (via secondary reports)

GPT-6.2 Astra launched on all three routes the same day: ChatGPT Work, Codex and the OpenAI API. This article gives the full model IDs, plans and allowances, API notes, Ultrafast multipliers and system-card thresholds — plus what it cannot do: no free tier, no weights, no fine-tuning.

Model IDs

Flagship: gpt-6.2-astra Everyday: gpt-6.2-sol Previous gen: gpt-6-astra / gpt-6.1-sol (superseded by 6.2 at each tier)

Call the API with gpt-6.2-astra directly; there are no per-cloud ID variants. The Responses API supports a multi-agent beta for delegating work to subagents, and discounted background-completion windows are in public beta for long tasks.

ChatGPT Work: Plus / Pro / Enterprise / Edu

Subscribers pick Astra inside ChatGPT Work, live on launch day. Included usage rose at launch: Plus and Standard Business seats went from 5–45 to 6–50 messages per 5 hours; Sol went from 15–160 to 17–180. Free users cannot use it — the easiest thing to get wrong about this launch: don't search for a "gpt-6.2 astra free" entry point, because there is no free tier.

Codex: live on launch day

Codex integrated Astra on launch day, suited to coding-agent work. Pair it with the Responses API's multi-agent beta and background-completion windows for long tasks that run in the background. Sol is on Codex too — everyday coding work is cheaper on Sol ($2/$10 vs $8/$40).

API notes

The Responses API rejects reasoning effort set to none — Astra is a reasoning model, and thinking cannot be switched off. Four effort tiers — Max / High / Medium / Low — with AA testing Max for the 62; start everyday work on High or Medium and reserve Max for the hardest tasks. Streaming, function calling and structured outputs are supported; tools include web search, file search, image generation, code interpreter, hosted shell, apply patch, Skills, computer use, MCP and tool search. Fine-tuning is not supported. Input is text and image, output is text; audio and video are not supported.

Ultrafast: 3× through Feb 28, 2027, then 6×

The Ultrafast decoding tier bills at 3× Standard rates through Feb 28, 2027 ($24/$120), then 6× from Mar 1, 2027 ($48/$240), Pacific time (OpenAI model page). Sol's Ultrafast already reaches the $200/month Pro plan. Start everyday work on Standard; reserve Ultrafast for the most latency-sensitive work, and confirm the price before the increase.

System-card thresholds: bio High, cybersecurity Critical

OpenAI's system card sets three thresholds: bio capability "High", cybersecurity "Critical", AI self-improvement "High" — OpenAI calls Astra the first released model at "High" for AI self-improvement. That means some high-risk requests will be blocked or routed elsewhere; the exact blocking behaviour is per the full system card, whose direct read is still pending — this site will add it once available.

What it cannot do

Not available to free users; no public weights, so no self-hosting route; no fine-tuning; no audio or video input or output; the Responses API rejects reasoning effort set to none. Requests above 272K input tokens bill the whole request at surcharge rates (input and cache rates double, output 1.5× — i.e. $16/$60), so budget long-document work at the surcharge first.

ModelBrief's take: Access itself has no barrier — one model ID, three live routes, four plans. The real barriers are two: free users are locked out, and the system-card thresholds mean some high-risk requests will be blocked. Try it first in ChatGPT Work or Codex, then decide whether to pay per token through the API; send batch work to Batch (50% of Standard).

Reference: OpenAI — GPT-6.2 Astra model page (read directly by this site, Nov 18, 2026) · Artificial Analysis — GPT-6.2 Astra model page (rewritten and organized by ModelBrief; facts per official sources and Artificial Analysis)