Hot AI News

GPT-6.2 Astra appears at $8 per million tokens, $2 below GPT-6 Astra

OpenAI's API models page now lists GPT-6.2 Astra and Sol with lower per-token prices than the GPT-6 Astra launch in September, but no announcement or release notes have appeared yet.

GPT-6.2 Astra appears at $8 per million tokens, $2 below GPT-6 Astra
Image: OpenAI

Two new entries, GPT-6.2 Astra and GPT-6.2 Sol, appeared on OpenAI’s API models page. Astra is listed at $8 per million input tokens and $40 per million output tokens. The page carries no launch date and no release notes.

This report rests on that page alone. OpenAI had not published a separate announcement or model card for GPT-6.2 in the sources checked, so what follows is what the listing shows, plus context from earlier coverage of GPT-6 Astra.

Output price per million tokens
Image: OpenAI, chart by Aiholics

What the listing shows for each model

The page now tells developers to start with GPT-6.2 Astra, which it calls the company’s flagship for complex reasoning and coding. Its model ID is gpt-6.2-astra. It is described as a frontier reasoning model for coding and professional work, with a 1.05M-token context window, up to 128K output tokens and a knowledge cutoff of May 18, 2026. Reasoning effort can be set to low, medium, high, xhigh or max. Listed tools are functions, web search, file search and computer use.

GPT-6.2 Sol (gpt-6.2-sol) is billed as the balance of intelligence and cost. It costs $2 per million input tokens and $10 per million output tokens, with the same context window, output cap, reasoning levels and tools. Its knowledge cutoff is later, June 1, 2026. OpenAI describes it as a reasoning model for text, images and tool use.

GPT-6.2 Astra appears at $8 per million tokens, $2 below GPT-6 Astra
Image: OpenAI

The rest of the flagship row is unchanged in structure: GPT-6.1 Terra at $1 and $5, and GPT-6 Luna at $0.10 and $0.50 for the most cost-sensitive, high-volume work. The page says all current models take text and image input, produce text, and are served through the Responses API and client SDKs. Unlike Terra and Luna, the two 6.2 models do not list a none reasoning setting.

Advertisements

Cheaper than the September launch prices

The prices are the most concrete news. According to a report on the GPT-6 Astra launch on September 3, 2026, that model was priced at $10 per million input tokens and $50 per million output tokens, with Sol at $4 and $20. Against those figures, GPT-6.2 Astra is 20% cheaper on both input and output, and GPT-6.2 Sol is 50% cheaper. Those earlier numbers come from a third-party write-up (VPS Ranking), not from OpenAI’s current page, and the Sol figure there may refer to an earlier Sol generation.

GPT-6.2 Astra appears at $8 per million tokens, $2 below GPT-6 Astra
Image: OpenAI

The same write-up said requests above 272K input tokens cost more and that Batch and Flex ran at half the standard rate. The current page does not mention long-context surcharges or discounts, so it is unclear whether those rules carry over. Anyone budgeting for the 1.05M window should check the pricing documentation before assuming the headline rate applies throughout.

What would confirm the picture

The listing leaves gaps. It does not say when the models went live, whether they are in ChatGPT, or whether they replace the earlier Astra and Sol. It also publishes no benchmarks. At the September launch, The Next Web reported that OpenAI’s published figures had not been independently reproduced and compared the model only with OpenAI’s own earlier one. The same report said OpenAI rated Astra at its “Critical” threshold for cyber capability, and that its reasoning was harder to monitor than Sol’s. Whether the 6.2 versions change that profile is not stated on the page.

Signs worth watching: a model card or a changelog entry from OpenAI, the contents of the Astra usage guide the page links to, and whether Azure and AWS Bedrock, which carried Astra at launch according to the VPS Ranking report, list the new IDs. Independent benchmark runs would show whether the lower price comes with a change in quality.

Other OpenAI releases are moving quickly. The model page also lists GPT-6 Cyber, announced on November 5, and the GPT-Image-2.6 pair introduced on November 4. Our earlier report on Copilot retiring older models shows how fast model lineups are turning over.

Sign Up for the Daily AI Pulse

One email a day. All the stories that matter.

Alex CarterAI News & Big Tech Correspondent

Alex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.

Leave a Reply

Your email address will not be published. Required fields are marked *

You may also like