OpenAI’s new GPT-6.1 Sol model gives you coding and agent skills close to GPT-6 Astra for about one-fifth of the price. It launched at DevDay on September 29, 2026, and you can use it today in the API, Codex, and ChatGPT Work. The catch: it isn’t in the regular ChatGPT chat window yet, and free users can’t touch it.
Quick Facts
- Released: September 29, 2026, at OpenAI DevDay
- API price: $2 input / $10 output per million tokens (cached input now $0.10)
- Context window: 1.05 million tokens in, 128,000 tokens out
- Where it works: OpenAI API, Codex, and ChatGPT Work on Plus, Pro, Business, Enterprise, and Edu
- Not included: ChatGPT Free and Go plans, and the standard chat window
What GPT-6.1 Sol Actually Is (and Why It Exists)
OpenAI now runs two families of GPT-6 models. Astra is the big, expensive flagship. Sol is the cheaper, faster workhorse meant for everyday coding and agent jobs.
GPT-6.1 Sol is the first upgrade to the Sol line. OpenAI pitches it as “near-Astra” intelligence for coding, computer use, and office-style work, according to OpenAI’s model docs. In plain terms, it’s built to do most of what Astra does without the Astra bill.
The timing matters too. Just before DevDay, OpenAI pulled the planned Astra upgrade after safety problems, which we covered in why GPT-6.1 Astra was cancelled. So Sol is the new model OpenAI actually shipped this cycle.
GPT-6.1 Sol Price: How Much Cheaper Is It?
This is the headline. The input and output prices stay the same as the older GPT-6 Sol, but OpenAI cut the cached input price in half. Here’s how it compares with Astra, based on figures reported by VentureBeat:
| Model | Input (per 1M) | Cached input | Output (per 1M) |
|---|---|---|---|
| GPT-6.1 Sol | $2 | $0.10 | $10 |
| GPT-6 Sol (previous) | $2 | $0.20 | $10 |
| GPT-6 Astra | $10 | — | $50 |
| GPT-6.1 Sol Ultrafast (coming) | $12 | — | $60 |
Cheaper caching is a big deal if you send the same long system prompt or codebase again and again. That’s exactly how most coding agents work, so your real-world savings can be larger than the sticker price suggests.
One warning on long prompts. Per DataCamp, any request over 272,000 input tokens is billed at 2x input and 1.5x output rates for the whole request. The 1-million-token window is real, but the cheap zone is much smaller.
How Close Does It Really Get to Astra?
OpenAI’s own numbers say “very close” on coding and “a step behind” on harder troubleshooting. These are company-reported benchmarks, so treat them as a starting point, not a final verdict.
Here are the results that matter most for everyday users:
- DeepSWE v1.1 (coding): matches GPT-6 Astra at roughly one-fifth the cost.
- OSWorld 2.0 (computer use): lands within 2.1 points of Astra.
- AutomationBench (business tasks): 2.2 points ahead of Claude Opus 5.5, per VentureBeat.
- Terminal-Bench Science: about $5.47 per task versus Astra’s $23.80.
The gap shows up in tougher, open-ended work. DataCamp reports a 15.5-point gap on TroubleshootingBench (47.96% vs 63.46%). If your job is debugging strange production failures, Astra still earns its price.
Where You Can Use GPT-6.1 Sol Right Now
Access depends on how you use OpenAI’s tools. The table below sums it up:
| Where | Who gets it | Notes |
|---|---|---|
| OpenAI API | All developers | Model ID: gpt-6.1-sol |
| Codex | Plus, Pro, Business, Enterprise, Edu | Desktop app, CLI, and IDE extension |
| ChatGPT Work | Plus, Pro, Business, Enterprise, Edu | Enterprise and Edu admins must turn it on |
| Regular ChatGPT chat | Nobody yet | Not available in standard conversations |
| ChatGPT Free and Go | No access | Not listed for these plans |
According to Fello AI, GitHub Copilot is also rolling it out gradually to Pro+, Max, Business, and Enterprise users. If you don’t see it yet, that’s likely why.
Developers should note one quirk. Per DataCamp, tool calling needs the Responses API, and the model has no “none” or “minimal” reasoning setting. The lowest option is “low,” so every answer does at least a little thinking.
Ultrafast: Paying Extra for Speed
DevDay also brought a new speed tier called Ultrafast. It pushes output up to 300 tokens per second, which OpenAI says is up to 8x faster in Codex and 6x faster in the API.
Speed costs six times the normal price for the same model. Astra Ultrafast is live now for Pro 500 and Enterprise users, while Sol Ultrafast was described as “coming soon” at launch. Most people won’t need it, but it makes sense for live coding sessions or support bots where every second of waiting hurts.
Frequently Asked Questions
Is GPT-6.1 Sol free in ChatGPT?
No. It’s limited to paid plans (Plus and above) inside ChatGPT Work and Codex. Free and Go users don’t get it, and it isn’t in the normal chat window yet.
How much does the GPT-6.1 Sol API cost?
It costs $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. Requests above 272,000 input tokens cost more.
Is GPT-6.1 Sol better than GPT-6 Astra?
Not overall. It matches Astra on several coding tests, but Astra still wins on hard troubleshooting. Sol wins on value, since it costs about one-fifth as much.
What is the GPT-6.1 Sol context window?
It accepts up to 1,050,000 input tokens and can write up to 128,000 output tokens. Its knowledge cutoff is April 30, 2026.
Our Take
If you build with OpenAI’s API or live inside Codex, switch your default to GPT-6.1 Sol and only reach for Astra when Sol gets stuck. The cached-input price cut alone makes it the smarter pick for agent work. Casual ChatGPT users can ignore this one for now — it isn’t in your chat window, and OpenAI’s DevDay 2026 launches were clearly aimed at developers and paying teams first.
Leave a Reply