Skip to content

GPT-6.2 Astra and GPT-6.2 Sol — OpenAI's New Frontier Pair, Same Bands ​

November 18, 2026

GPT-6.2 Astra and GPT-6.2 Sol are in the model picker, in the seats GPT-6 Astra and GPT-6.1 Sol held. Astra is OpenAI's most capable model for the hardest end-to-end work; Sol is the everyday frontier for coding and long agent runs, and stays on Recommended. Both read 1.05M tokens of context, write back up to 128K, take text and images, and run the full five effort levels from low through max. Bands do not move: Astra stays Apex 18×, Sol stays Premium 6×. Open a GPT-6 Astra or GPT-6.1 Sol chat and the next turn runs on its 6.2 replacement — you do not re-pin.

What you can do ​

  • Pin GPT-6.2 Sol from Recommended. It keeps Sol's seat on the shortlist, one click from a new chat.
  • Pin GPT-6.2 Astra from Closed SOTA when the job deserves OpenAI's flagship — the hardest end-to-end reasoning, research, and document work.
  • Start from the new-chat card. It names GPT-6.2 Astra and GPT-6.2 Sol and opens the picker on the Astra row, with Sol directly beneath it.
  • Put long material and screenshots in the same turn. Both read up to 1.05M tokens and take text and image input.
  • Get the finished document back, not an outline. Both write up to 128K tokens in one reply — a full memo or board pack in a single turn.
  • Choose how hard they think. Fast and Medium map to low and medium effort; a pinned Beast asks for max, the top of all five levels. On Sol, effort, verbosity, and pro reasoning also scale with how hard the turn looks.
  • Keep working in an existing thread. GPT-6 Astra chats continue on GPT-6.2 Astra, and GPT-6.1 Sol chats on GPT-6.2 Sol, with saved model preferences rolling forward the same way.

Where this shows up ​

You have a data room to get through before Thursday's board call — the contracts, the financial model, three years of filings. Pin GPT-6.2 Astra in Beast, put all of it in one turn, and get the diligence memo with the numbers that do not hold up, not a pile of summaries.

Legal sent a redline ninety minutes before the deal call. Pin GPT-6.2 Sol, attach both versions, and get the clauses that changed and the one to push back on first.

Finance sent scanned invoices and you have the dashboard open in another tab. Pin GPT-6.2 Sol, attach the scans and the screenshots together, and ask where the two disagree.

You are carrying a research brief that already outgrew the model you started on. Open the same thread — it is on 6.2 now — and keep going without re-pinning or losing the history.

Try it ​

  • Pin GPT-6.2 Astra in Beast, attach the data room, then:

    "Write the diligence memo for Thursday's board: what the deal is, the three numbers that do not hold up, and whether we proceed."

  • Pin GPT-6.2 Sol, attach the five supplier contracts, then:

    "List every clause that differs from our standard terms, with the page it is on, and tell me which three to renegotiate first."

  • Pin GPT-6.2 Sol, attach the scanned invoices and the dashboard screenshots, then:

    "Pull every figure off the scans, match each one to the dashboard, and tell me where the two disagree and which page the number came from."

  • Open the GPT-6.1 Sol thread — it is now pinned to GPT-6.2 Sol — then:

    "Pick the research brief up where we left it and turn what we have into the final client memo."

Heads up ​

  • Same bands as before. Astra is Apex 18× and Sol is Premium 6×, exactly where their predecessors sat. Your plan quota and token balance do not change.
  • GPT-6 Astra and GPT-6.1 Sol are retired from the picker, Auto, and Swarm. Sessions and saved preferences pinned to either one roll forward automatically, and stored pins update when you open the chat or send the next message. GPT-6 Sol and GPT-5.6 Sol pins now land on GPT-6.2 Sol too.
  • Astra stays a manual pick. At Apex 18× it is kept out of Auto and Swarm, as GPT-6 Astra was — pin it yourself when you want it.
  • Sol works in Auto and Swarm. Auto can pick it on Pro Max and Black, where quality 68 clears the paid Auto floor of 60, and it can take a Swarm step. Pro's Auto stays inside its curated pool, so pin Sol there when you want it every turn.
  • GPT-6.1 Terra and GPT-6 Luna are unchanged — same seats, same bands. Argus heartbeats still default to GPT-6 Luna.
  • No harness numbers yet. Can it Run Alfrada? will not show GPT-6.2 Astra or Sol until each has a full run.

Built for Alfrada OS.