What Is GPT-6.1 Sol?

GPT-6.1 Sol is OpenAI’s newest Sol model, launched at DevDay on September 29, 2026, with the API model ID gpt-6.1-sol. It costs $2 per million input tokens and $10 per million output tokens, the same list price as GPT-6 Sol, but cached input drops from $0.20 to $0.10. It takes up to 922,000 input tokens in a 1,050,000-token context window, writes up to 128,000 tokens, and has an April 30, 2026 knowledge cutoff. OpenAI pitches it as “near-Astra intelligence” at one-fifth of GPT-6 Astra’s standard token prices.

This explainer covers availability, pricing, specs, OpenAI’s benchmark claims, and when GPT-6 Astra is still the better pick. For everything else OpenAI shipped that day, see the DevDay 2026 roundup; for the model it succeeds, read what GPT-6 Sol is. You can send your first gpt-6.1-sol request from Apidog and keep it as a saved test.

GPT-6.1 Sol at a glance

Source: the GPT-6.1 Sol model page. The window and input limit match what the GPT-6 Sol model page lists today, so context size isn’t the upgrade.

Where you can use GPT-6.1 Sol

OpenAI’s launch post makes it available “to all Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex,” and adds that it “is not yet available in Chat.” Free and Go aren’t on the list, and OpenAI’s models docs say Enterprise and Edu keep it off until an administrator enables it. In the API it’s gpt-6.1-sol on Chat Completions, Responses and Batch, billed per token, and OpenRouter serves it as openai/gpt-6.1-sol at the same list rates on its OpenAI endpoint.

There’s no free API tier. The GPT-6.1 Sol free guide ranks the cheapest routes and names a free alternative (a different model).

GPT-6.1 Sol pricing

All prices are per million tokens, from OpenAI’s pricing page, for prompts up to 272K input tokens:

Ultrafast for GPT-6.1 Sol is “coming soon,” with no price yet. Three things stand out in the table.

Per token, the upgrade is free. Input and output list prices match GPT-6 Sol exactly. The only rate that moved is cached input, which OpenAI prices at “$0.10 per million tokens, 95% less than standard input pricing and 50% less than GPT-6 Sol’s cached input pricing.” Reuse a long system prompt, tool schemas or repo context, and that line of your bill halves. Output token counts can shift with a new model, so measure before you promise savings. The GPT-6 prompt caching guide shows how to structure prompts that hit the cache.

Long prompts cost more. Past 272K input tokens, the model page says the whole request bills at 2x the input and cache rates and 1.5x the output rate: $4 input, $0.20 cached and $15 output on Standard. You can send 922K tokens, but you pay double on input to do it.

Astra costs five times as much. GPT-6 Astra lists at $10/$50, five times 6.1 Sol’s standard rates, which is where OpenAI’s “one-fifth” framing comes from. Astra is also the only model with Ultrafast broadly available today, at 6x its Standard price.

What changed from GPT-6 Sol

The GPT-6 Sol model page now says “See GPT-6.1 Sol for the newer Sol model.” Switching costs you two things. Any request that sends reasoning: {"effort": "none"} needs a new value: move it to low, the lowest level 6.1 Sol accepts, and re-run your evals, because low still reasons and reasoning tokens bill as output. The knowledge cutoff also moves forward ten days. The GPT-6.1 Sol API guide walks through the migration with request examples.

GPT-6.1 Sol benchmarks: what OpenAI claims

Every number in this section comes from OpenAI’s launch post: OpenAI ran its own models and says competitor results come from publicly available reports. OpenAI publishes the raw scores as charts; the deltas below are the ones it states in text. Costs are cost per task at the stated effort.

Read the table with three caveats:

• The opponents are Opus 5.5 and Fable 5.1, not Claude Sonnet 5.5, which launched a day earlier at the same $2/$10. No vendor has published a 6.1 Sol vs Sonnet 5.5 head-to-head; the GPT-6.1 Sol vs Claude Sonnet 5.5 guide shows how to run one on your own prompts. On AutomationBench, OpenAI also notes the Fable 5.1 cost point omits fallbacks, which occurred on about 40% of tasks.
• The pitch is cost per task, not peak score. Most rows say “matches” or “approaches” Astra at a fraction of the cost. Strong for budgets, weaker if you need the top score.
• Factuality was measured on conversations users flagged for errors, not typical usage.

Astra still wins one headline test. On Terminal-Bench Science, OpenAI says Astra still scores highest, at 68.1%, and recommends Astra for the hardest science work.

When GPT-6 Astra is still the right call

For most coding and agent workloads that ran on GPT-6 Sol, 6.1 Sol is a like-for-like swap at the same price. Keep Astra when:

• The task is the hardest science or terminal work. That’s OpenAI’s own guidance after Terminal-Bench Science.
• You need Ultrafast today. Astra Ultrafast is live in the API and in ChatGPT Work and Codex on Pro 500 and Enterprise. 6.1 Sol’s is “coming soon.”
• You’re building on the Agents API’s computer-use tool. Its docs name gpt-6-astra only. The 6.1 Sol model page lists computer_use as a tool, so check the Agents API docs before you move an agent over.
• The last couple of points matter more than a 5x cost gap. On OSWorld and factuality, Astra stays about 2 pp ahead.

In the other direction, GPT-6 Luna at $0.10/$0.50 covers high-volume classification and extraction where Sol-level reasoning is overkill. The Luna vs Sol vs Astra comparison works through cost per task by tier, and the GPT-6 Astra API guide covers Astra’s parameters.

Call GPT-6.1 Sol and check what it costs

Here’s a first request on the Responses API, with the key read from an environment variable:

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": "low"},
    "input": "List three ways to make a webhook receiver idempotent."
  }'

The response carries a usage object with input_tokens, output_tokens and output_tokens_details.reasoning_tokens. OpenAI’s reasoning guide says reasoning tokens are billed as output, so effort changes your bill even when the visible answer is the same length. Cache hits show up in input_tokens_details.cached_tokens.

To turn that into a repeatable check in Apidog:

1. Create an environment with OPENAI_API_KEY and MODEL_ID set to gpt-6.1-sol. Send Bearer {{OPENAI_API_KEY}} in the Authorization header and "model": "{{MODEL_ID}}" in the body.
2. Save the request as a POST to https://api.openai.com/v1/responses.
3. Add assertions: status is 200, $.usage.output_tokens is greater than 0, and $.usage.output_tokens_details.reasoning_tokens stays below a ceiling you pick for this prompt.
4. Put a fixed prefix of at least 1,024 tokens (the minimum cacheable length in the prompt caching guide) at the start of input, send twice, and assert $.usage.input_tokens_details.cached_tokens is greater than 0 on the second run.
5. Clone the environment with MODEL_ID set to gpt-6-sol and run the same request in both. Same prompt, two models, usage side by side.

FAQ

Is GPT-6.1 Sol free? No. It’s in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu, and the API has no free tier. The GPT-6.1 Sol free guide covers the lowest-cost routes.

What is the GPT-6.1 Sol model ID? gpt-6.1-sol, a single snapshot. On OpenRouter it’s openai/gpt-6.1-sol.

Does GPT-6.1 Sol support none reasoning effort? No. It accepts low, medium (the default), high, xhigh and max. Requests that sent none to GPT-6 Sol need a new value.

Is GPT-6.1 Sol better than GPT-6 Astra? Not across the board. OpenAI says it matches or approaches Astra on several benchmarks at one-fifth to one-seventh the cost per task, but Astra still leads Terminal-Bench Science at 68.1%.

Next step

If you run GPT-6 Sol today, change the model ID, replace any none effort with low, and compare usage on a handful of real prompts before you switch production traffic. Download Apidog to keep both runs as saved requests with assertions, then follow the GPT-6.1 Sol API guide for the full migration checklist.