OpenAI releases GPT-6 Sol and GPT-6 Luna, and halves the price of both tiers

OpenAI's API changelog recorded on 22 September 2026 that it [released GPT-6 Sol, model id gpt-6-sol, and GPT-6 Luna, model id gpt-6-luna] . [Both are reasoning models that accept text and image inputs and generate text through the Responses and Chat Completions APIs] , and the whole GPT-5.6 family left the catalogue's flagship rows the same day.

What changed

Two models, and a price list that reads lower at every tier. [The prices the announcement states are standard pricing per 1M tokens for prompts with up to 272K input tokens] : [GPT-6 Sol is announced at $2 input, $0.20 cached input, and $10 output] , and [GPT-6 Luna at $0.10 input, $0.01 cached input, and $0.50 output] . The pricing page's own table adds the long-context column: [gpt-6-sol is $2.00, $0.20, $2.50 and $10.00 at short context and $4.00, $0.40, $5.00 and $15.00 at long] , and [gpt-6-luna $0.10, $0.01, $0.125 and $0.50, then $0.20, $0.02, $0.25 and $0.75] .

The catalogue now sorts three models rather than three of a different family. [It tells a reader to start with GPT-6 Astra for complex reasoning and coding, choose GPT-6 Sol to balance intelligence and cost, or use GPT-6 Luna for cost-sensitive, high-volume workloads] . [Astra, model id gpt-6-astra, is described as the most capable model, built for the hardest end-to-end work] ; [Sol as built to power complex coding and agentic workflows] ; [Luna as the most efficient model for focused, high-volume tasks] . [Each of the three lists a maximum output of 128K tokens and a context window of 1.05M] .

What it was before

The names that stood in those rows until this week were GPT-5.6 Sol, Terra and Luna, and this site quoted them: two of our guides cite the model catalogue for the row that named GPT-5.6 Luna, and since the page changed those guides say "verification overdue" on their own pages — What Tool Use Is and What Prompt Caching Is. That is the daily source check doing its job rather than a fault: the quotation the guide was signed against is no longer on the page it cites, so its line changes state until a person reads the new page and signs it again.

The old family has not been struck from the price list, though. [The same pricing page still states that GPT-5.6 Sol's promotional pricing is available at least through November 21, 2026] — a sentence about the previous flagship, still there under the new one.

What it means for you

If you call the API with a model id written into your code, nothing broke this morning and nothing will break silently: an id is a string OpenAI keeps serving or stops serving, and the page above says the older tier is still priced. But the arithmetic under your bill has changed, and in your favour if you move.

Check three things. The id you send, in every place it is written — application code, evaluation scripts, notebooks, a colleague's cron job. What you actually pay for, because the cached-input column is where the difference compounds: at $0.01 per million tokens, GPT-6 Luna's cached input is the cheapest number on that table, and it rewards a prompt whose prefix does not move. What your prompts are worth re-measuring: the announced prices apply to prompts up to 272K input tokens, and a workload that sits above that line is priced by the long-context column, not the headline.

If you keep a note of what your stack costs per thousand requests, this is the week to re-run it. The names changed, the tiers did not, and the numbers went one direction.

The source

OpenAI's API changelog, the entry of 22 September 2026, with the model catalogue and the pricing page as they read that day.