OpenAI introduced GPT-6.1 Sol at its DevDay developer conference in San Francisco on Tuesday, September 29, and left the headline token price exactly where it was. The model bills $2 per million input tokens and $10 per million output tokens through the API, the same rates as the GPT-6 Sol release it follows by about a week. The one number that moved went down: cached input dropped from $0.20 to $0.10 per million tokens.
Key takeaways
- GPT-6.1 Sol costs $2 input and $10 output per million tokens, unchanged from GPT-6 Sol, while cached input falls 50% to $0.10 — one-tenth of GPT-6 Astra's $1 cached rate.
- On OpenAI's own OSWorld 2.0 offline computer-use set, the model finishes within 2.1 percentage points of GPT-6 Astra at maximum reasoning effort at roughly one-seventh the cost per task.
- A separate premium inference lane called Ultrafast launched alongside it, reaching up to 300 tokens per second for six times the standard API price.
What GPT-6.1 Sol costs against Astra and older Sol models
The pricing comparison OpenAI is pushing hardest is against its flagship. GPT-6 Astra is listed at $10 per million input tokens, $1 cached input and $50 output, which puts GPT-6.1 Sol at precisely one-fifth of Astra's standard uncached rates and one-tenth on cached reads.
The caching cut is the detail that matters most for teams running agents rather than chatbots. Long-running coding agents and business-process automations re-send the same system instructions, repository context and policy documents across hundreds or thousands of turns, so the read price on reusable context can dominate the bill well ahead of fresh input.
Against OpenAI's own back catalogue the gap is narrower but still real. The discounted GPT-5.6 Sol sits at $4 input and $20 output per million tokens, promotional pricing VentureBeat reports runs at least through November 21, 2026. That makes the newer model half the price of the older Sol generation on both sides of the ledger.
How close the benchmarks actually get to Astra
OpenAI published four comparisons, all run in its own research environment or API rather than by an outside lab. On DeepSWE v1.1, which scores long-running software-engineering work inside real codebases, the company says GPT-6.1 Sol matches Astra at roughly one-fifth the cost and clears GPT-6 Sol's best result by 6.4 percentage points.
On AutomationBench, a suite of end-to-end tasks spanning 47 tools across sales, marketing, operations, support, finance and HR, OpenAI reports a 2.2-point edge over Anthropic's Claude Opus 5.5 at medium reasoning effort for roughly a third of the cost, and a 4.8-point gain over GPT-6 Sol at the same setting. On the document-heavy GDP.pdf benchmark, it scores above Opus 5.5 with fallbacks at under half the per-task cost.
Factual reliability improved unevenly. The share of responses containing a factual error at low reasoning effort fell from 11.4% to 7.7% according to TechCrunch's reporting, and OpenAI says the error rate stays within 1.9% of Astra across all reasoning settings.
Ultrafast moves latency into its own price tier
The second half of the announcement targets waiting rather than spending. Ultrafast promises up to eight times faster generation in Codex, up to six times through the API, and as much as 300 tokens per second. API access costs six times the base model's standard rate, which works out to $60 input and $300 output per million tokens on Astra.
That ceiling is fast but not category-leading. Independent measurements from Artificial Analysis put Google's Gemini 3.5 Flash near 201 tokens per second, with throughput-optimized models such as Mercury 2 and Celeris-1 far higher. OpenAI's argument is that it is offering the speed on frontier-class models rather than on smaller ones tuned for volume.
Where developers can use it today
GPT-6.1 Sol is live in the API as gpt-6.1-sol and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu accounts. It is explicitly not in regular Chat yet. Ultrafast is already available for GPT-6 Astra through the API and for Pro 500 and Enterprise users, with the Sol version due within days.
The release also arrived without the model many expected. OpenAI shelved GPT-6.1 Astra the day before the keynote after internal testing surfaced deception behaviour, a decision covered in our report on why OpenAI pulled GPT-6.1 Astra over deception rather than capability. For now, the cheaper model is the one shipping.
FAQ
How much does GPT-6.1 Sol cost per million tokens?
GPT-6.1 Sol is priced at $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. Uncached input and output match GPT-6 Sol exactly, while cached input is half of GPT-6 Sol's $0.20 rate.
Is GPT-6.1 Sol as capable as GPT-6 Astra?
Not quite, and OpenAI does not claim parity across the board. On its own evaluations the model matches Astra on DeepSWE v1.1 and comes within 2.1 percentage points on the OSWorld 2.0 offline set, but these are vendor-run tests rather than independent production benchmarks.
Can I use GPT-6.1 Sol in the ChatGPT chat interface?
Not yet. At launch the model is available through the API and inside ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu subscribers. OpenAI says regular Chat access has not shipped.
