OpenAI releases GPT-6.1 Sol, at GPT-6 Sol's input and output prices and half its cached-input price

OpenAI released GPT-6.1 Sol, model id gpt-6.1-sol, on 29 September 2026, for complex coding and professional work at a lower cost than GPT-6 Astra. Its announced input and output prices are GPT-6 Sol's; its cached input costs half as much. The model catalogue now names it, not GPT-6 Sol, as the model that balances intelligence and cost. If you callgpt-6-sol or gpt-6-astra today, this is a model to measure — with two things in your code to check before you switch.

What changed

The changelog states the price: standard pricing per 1M tokens, for prompts with up to 272K input tokens, is $2 input, $0.10 cached input, $2.50 cache write and $10 output. The model page puts the cached rate another way: cached input tokens cost 5% of the uncached input rate. Past 272K input tokens the whole request is repriced — 2x the input and cache rates and 1.5x output. The pricing page's flagship table carries both columns:gpt-6.1-sol at $2.00, $0.10, $2.50 and $10.00 at short context, then $4.00, $0.20, $5.00 and $15.00 at long, between gpt-6-astra and gpt-6-luna.

The model page describes GPT-6.1 Sol as near-Astra performance at a lower cost for complex coding, computer use and professional work , and lists a 1,050,000-token context window, 128,000 maximum output tokens and a knowledge cutoff of 30 April 2026. It takes text and images in and gives text out; audio and video are not supported.

Two details decide whether your existing calls carry over. reasoning.effort accepts low, medium (the default), high, xhigh and max — none and minimalare not supported. And tool calling goes through the Responses API: Chat Completions works, but without tool calling. The model also supports Multi-agent in beta, which lets it delegate work to subagents within one Responses API request.

What it was before

A week earlier, on 22 September, OpenAI released GPT-6 Sol (gpt-6-sol) and GPT-6 Luna , and we covered it inOpenAI releases GPT-6 Sol and GPT-6 Luna. As that item recorded, the catalogue then told you to choose GPT-6 Sol to balance intelligence and cost. GPT-6 Sol was announced at $2 input, $0.20 cached input and $10 output : the same input and output as GPT-6.1 Sol, and twice its cached input.

Today the catalogue's Sol row reads GPT-6.1 Sol, model id gpt-6.1-sol, and the pricing page's flagship table has three rows —gpt-6-astra, gpt-6.1-sol, gpt-6-luna — with no row for gpt-6-sol. Neither page says whethergpt-6-solstops being served, or when; the changelog points to the deprecations page for that.

What it means for you

If you are on gpt-6-sol, the announced input and output prices stay where they are and your cached input halves, from $0.20 to $0.10 per million tokens. The more of your prompt that is a fixed prefix read from cache, the more of your bill that halving reaches. Our guideWhat Prompt Caching Isexplains how a prefix gets cached and already quotes GPT-6.1 Sol's read rate — the same 5% the model page states.

If you are on gpt-6-astra, the table puts the difference at a factor of five on input ($10.00 against $2.00) and output ($50.00 against $10.00), and ten on cached input ($1.00 against $0.10). That gap is the "lower cost" the changelog names, and the model page's own advice is to compare the two on your tasks to weigh quality against cost. Run your evaluation set on both before you move production traffic.

Before you switch, check four things.

• The id string. Send gpt-6.1-sol, exactly — the model page names it as the id that selects this model. Search every place an id is written: application code, evaluation scripts, notebooks, scheduled jobs.
• Your tool calls.If your agent calls tools through Chat Completions, it has to move to the Responses API first; on Chat Completions, GPT-6.1 Sol runs without tool calling.
• Your reasoning setting. A request that sends reasoning.effort as none or minimalis outside what this model supports; pick low or above.
• Your rate limits.They depend on your usage tier: 5,000 requests and 1,000,000 tokens per minute at Build, 10,000 and 4,000,000 at Launch, 15,000 and 40,000,000 at Grow.

If your prompts run past 272K input tokens, price them with the long-context column, not the headline: the multiplier applies to the full request.

The source

OpenAI's API changelog, the entry of 29 September 2026, with the GPT-6.1 Sol model page, the model catalogue and the pricing page.