GPT-6.1 Sol shipped on September 29, 2026, and the GPT-6.1 Sol pricing is the headline: $2 per million input tokens and $10 per million output, a fifth of what GPT-6 Astra charges. If you own an API bill, re-price your workload. If you administer GitHub Copilot, the model is on for your organisation unless you've already told Copilot otherwise, so there's a decision waiting for you.
For the wider run of releases, see AI model releases and pricing: what changes for engineering teams.
What GPT-6.1 Sol is and what changed from GPT-6 Sol
OpenAI's API changelog describes it as a model for complex coding and professional work at a lower cost than GPT-6 Astra. Its guidance page says to use it when you want "near-Astra performance at a lower cost."
Compared with GPT-6 Sol, almost nothing moved on the price sheet. Input is still $2, output still $10, context still 1,050,000 tokens, max output still 128,000. The one price change is cached input, which halves from $0.20 to $0.10. The knowledge cutoff moves from April 20 to April 30, 2026.
| Per 1M tokens | GPT-6 Astra | GPT-6 Sol | GPT-6.1 Sol |
|---|---|---|---|
| Input | $10 | $2 | $2 |
| Cached input | $1 | $0.20 | $0.10 |
| Cache write | not listed | $2.50 | $2.50 |
| Output | $50 | $10 | $10 |
| Knowledge cutoff | April 30, 2026 | April 20, 2026 | April 30, 2026 |
Astra's figures come from our GPT-6 Astra release breakdown, which also covers the Responses API migration, so we won't repeat it here. If you read our earlier Claude Sonnet 5.5's price parity with GPT-6 Sol comparison, note that its $0.20 cached figure is GPT-6 Sol's rate. GPT-6.1 Sol's is $0.10.
What it costs: the cost math for a team
Against Astra the ratios are clean: 5x cheaper on input ($10 to $2), 5x on output ($50 to $10), and 10x on cached input ($1 to $0.10). Whether you see 5x or 10x on a real bill depends on how much of your traffic hits the cache.
Here's an illustrative month, not a benchmark. Assume 100M uncached input tokens, 400M cached input tokens and 20M output tokens.
- Astra: 100 x $10 = $1,000, plus 400 x $1 = $400, plus 20 x $50 = $1,000. Total $2,400.
- GPT-6 Sol: 100 x $2 = $200, plus 400 x $0.20 = $80, plus 20 x $10 = $200. Total $480.
- GPT-6.1 Sol: 100 x $2 = $200, plus 400 x $0.10 = $40, plus 20 x $10 = $200. Total $440.
That's $2,400 against $440, about 5.5x, and $40 a month saved against GPT-6 Sol, roughly 8%. Cache-heavy agent workloads gain the most from the cached cut. A chat product with little caching will barely notice the difference from Sol.
Three caveats from OpenAI's model page change the arithmetic:
- Prompts over 272K input tokens are priced at 2x on input and cache rates and 1.5x on output, for the full request, not just the tokens past the line. Long-context work costs $4 input and $15 output per million.
- Cache writes cost 1.25x the input rate, so $2.50.
- Batch gets 50% off standard pricing.
And "near-Astra" is OpenAI's framing, not a measured parity. Don't move a workload because of that phrase. Run your own tasks through both models and compare cost per accepted result, since a model that needs a second attempt erases a lot of a 5x saving.
For a per-seat view of the same question, see what AI coding tools cost per developer.
Multi-agent beta in the Responses API
The changelog says GPT-6.1 Sol supports Multi-agent in beta: the model can delegate work to subagents inside a single Responses API request. It's a beta, so treat the interface as something that can change.
The cost risk is plain even without numbers. Subagents do their own work, and that work is more tokens. A cheap per-token rate says little about a task that fans out into several parallel threads. Watch cost per completed task, not cost per token, and set a budget ceiling before you turn it on in anything automated.
Copilot: the on-by-default decision for admins
GitHub's announcement says GPT-6.1 Sol is generally available in Copilot for Pro+, Max, Business and Enterprise users. Copilot Pro without the plus doesn't get it. GitHub also says that "in early testing, it reliably completed tasks while using noticeably fewer tokens and steps than earlier models in the GPT-6 and GPT-5.6 families." That's GitHub's claim, and we haven't measured it.
For Business and Enterprise, admins manage access through the model policy in Copilot settings. The catch: new models are enabled unless an administrator has turned off the global default or explicitly disabled that model. It's billed at provider list pricing under usage-based billing, so a default-on model changes your spend whether or not anyone asked for it. Copilot's usage-based billing rules explain how that lands on an invoice.
You have three options:
- Leave it on. Reasonable if you've budgeted for usage-based billing and want developers on the current model.
- Disable GPT-6.1 Sol for specific organisations, such as one where cost tracking isn't set up yet.
- Turn the global default off. That stops this model, but every future model is then off too until someone enables it, so you've created a standing chore.
For most teams the second option is the right call. It handles the cost worry without making you the bottleneck for every release.
Codex CLI default and other things that change
The Codex changelog lists Codex CLI 0.159.1 adding GPT-6.1 Sol as the default model in the bundled catalog and the Amazon Bedrock Mantle and Runtime catalogs. If you run Codex unpinned, your default just changed under you. Pin a model if you need stable results.
OpenAI's guidance page says GPT-6.1 Sol doesn't support none reasoning effort, so use low. OpenAI advises keeping your current effort setting first and then testing one level lower.
There's also a fix in the API changelog: a bug in image encoding had degraded image understanding in GPT-6 Sol and GPT-6 Luna. Re-run any vision evals you ran on Sol or Luna before the September 25 fix.
FAQ
How much does GPT-6.1 Sol cost?
$2 per million input tokens, $0.10 cached input, $2.50 cache write and $10 per million output, for prompts up to 272K input tokens. Above that, input and cache rates double and output rises 1.5x for the full request. Batch is 50% off.
Is GPT-6.1 Sol cheaper than GPT-6 Sol?
Only on cached input, which drops from $0.20 to $0.10 per million tokens. Input, output, context window and max output are unchanged. The knowledge cutoff moves from April 20 to April 30, 2026.
Is GPT-6.1 Sol on by default in GitHub Copilot?
For Business and Enterprise, yes, unless an admin has turned off the global default or explicitly disabled the model in the model policy.
Which Copilot plans get GPT-6.1 Sol?
Copilot Pro+, Max, Business and Enterprise.
Is GPT-6.1 Sol the Codex default?
Yes, from Codex CLI 0.159.1, released September 29, 2026.