GPT-6.1 Sol

GPT-6.1 Sol is the September 29, 2026 upgrade to GPT-6 Sol for coding, computer use and professional work. Its API ID is gpt-6.1-sol. Use it in ChatGPT Work, Codex or the API. This page brings together the settings and decisions needed to use it; the GPT-6.1 overview explains the series and launch.

GPT-6.1 Sol pricing

Standard API rates, in USD per million tokens:

Token category	Up to 272K input	Above 272K input
Uncached input	$2.00	$4.00
Cached input	$0.10	$0.20
Cache writes	$2.50	$5.00
Output	$10.00	$15.00

The long-context rates apply to the whole request when input exceeds 272,000 tokens. Fast costs twice Standard; Batch and Flex cost half Standard. Regional processing adds 10% where available, and Fast is unavailable with EU data residency. Reasoning tokens contribute to billable output.

For a short Standard request with 10,000 uncached input tokens and 2,000 billable output tokens, the text-token cost is $0.04. If a later request reuses 8,000 cached input tokens and adds 2,000 uncached input tokens, with the same output, the cost is $0.0248. These calculations exclude cache creation, tools and regional charges. Keep the cache category visible in your logs: writing a prefix and reading it later have different rates.

API token billing is separate from a ChatGPT subscription allowance. For a repeated workflow, record actual token usage and retry counts before projecting a monthly bill. The prompt caching guide explains the read/write accounting; select a model in the tool only when its displayed rates match the reference you need.

Context window and reasoning settings

Sol has a 1,050,000-token context window and 128,000 maximum output tokens. It accepts text and images and produces text. Native audio, video and fine-tuning are unsupported. Supported reasoning efforts are low, medium, high, xhigh and max; medium is the default. Neither none nor minimal is supported.

Start with medium on a small representative task. Save cases where the answer misses a requirement, then test a higher effort on those cases. For document work, include the needed source sections and reserve room for the answer. A large context limit gives you room to work; a focused input still makes citations and review easier.

How to access GPT-6.1 Sol

In ChatGPT, open Work or Codex and select GPT-6.1 Sol. The launch covers Plus, Pro, Business, Enterprise and Edu; the release notes describe a staged rollout. If the option is missing, update your client and check the account and workspace model permissions. Sol is not available in ordinary Chat at launch.

For the API, use an API project with billing and model access, then set model to gpt-6.1-sol. A ChatGPT subscription does not supply API credits. The API setup page covers project keys and request setup.

In Work and Codex, the Standard credit rate is 50 input, 2.5 cached-input and 250 output credits per million tokens, with no separate cache-write charge. Fast consumes included subscription usage at 2.5x Standard, while purchased-credit usage uses 2x. These are different from the API-dollar rates above. Pro $200’s allowance transition and the other Pro tiers are covered in the plan and allowance guide.

Ultrafast was announced as coming in the days after launch. Check the current picker and official model reference before planning around that tier; the rates above describe the published Standard, Fast, Batch and Flex settings.

A first gpt-6.1-sol API request

Use Responses for tool calling. Chat Completions supports this model without tool calling. This text-only request provides a small starting test:

curl https://api.openai.com/v1/responses \
  -H "Authorization: Bearer $OPENAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-6.1-sol",
    "reasoning": {"effort": "medium"},
    "max_output_tokens": 1500,
    "input": "Review this proposed change: paginate a list of 23 items in pages of 10. List boundary cases and an acceptance test before suggesting code."
  }'

Check the returned model, completion status and usage fields. The output budget includes reasoning, so an incomplete response may need a larger budget. Then replace the sample task with a real input and explicit acceptance criteria. A repository task should include the observed failure, relevant paths and test command; a document task should specify source identifiers and the answer format.

GPT-6.1 Sol vs GPT-6 Sol

The two versions use different IDs: gpt-6.1-sol and gpt-6-sol. Cached input falls from the original Sol reference’s $0.20 to $0.10 per million tokens at Standard short-context rates. The input and output headline rates remain $2 and $10.

Review endpoint and effort settings during migration. An older request using none, or Chat Completions with tools, needs adjustment for 6.1 Sol. Compare results on the same saved inputs and count successful completions, corrections and retries. Keep the original Sol model page as the prior-version reference. The pagination-review example was run with the original model and supplies a task you can adapt for a new comparison.

When to use Sol or Astra

OpenAI positions 6.1 Sol near Astra on selected evaluations at lower token prices. Test that tradeoff on your workload: run the same task with the same sources, tools and review rubric, then compare accepted results and total cost. Send the difficult failures to GPT-6 Astra and measure whether it resolves them.

For a dated account of the release, read the GPT-6.1 Sol launch news. Return to GPT-6.1 to browse the version overview.