GPT-6.1 Sol launches with near-flagship performance at a fifth of the price
GPT-6.1 Sol brings near-Astra benchmark results to the API and ChatGPT Work at $2/M input and $0.10/M cached tokens.
OpenAI announced GPT-6.1 Sol on 29 September 2026 at its DevDay 2026 event. The headline claim is straightforward: benchmark performance that nearly matches GPT-6 Astra, the company’s most capable model, at roughly one-fifth of Astra’s API input and output token prices.
That is a meaningful gap to close. GPT-6 Astra costs $10 per million input tokens and $50 per million output tokens. GPT-6.1 Sol comes in at $2 per million input tokens and $10 per million output tokens. For teams running agentic workloads or high-volume coding pipelines, that difference compounds quickly.
Where it fits in the model family
OpenAI’s GPT-6 series currently has three tiers. GPT-6 Astra sits at the top as the flagship, priced at $10 input / $50 output / $1 cached. GPT-6 Luna sits at the bottom for fast, efficient everyday tasks at $0.10 input / $0.50 output / $0.01 cached. GPT-6.1 Sol occupies the middle: capable enough to approach Astra on several benchmarks, priced much closer to Luna.
It also replaces GPT-6 Sol as the mid-tier option, with identical standard input and output pricing but a significantly cheaper cached rate. GPT-6 Sol charged $0.20 per million cached input tokens. GPT-6.1 Sol halves that to $0.10.
What the benchmarks actually show
OpenAI published results across three areas: software engineering, computer use, and professional automation.
On DeepSWE v1.1, which tests real-codebase software engineering tasks, GPT-6.1 Sol matches GPT-6 Astra while outperforming GPT-6 Sol by 6.4 percentage points at lower reasoning effort and cost. On OSWorld 2.0’s offline set, which evaluates computer use, GPT-6.1 Sol scores 71.4% versus Astra’s 73.5%, both at maximum reasoning effort, at roughly one-seventh of Astra’s cost per task. On AutomationBench, the model scores 31.7% at medium reasoning effort, up 4.8 percentage points from GPT-6 Sol at the same setting.
Factual accuracy also improved. At low reasoning effort, 7.7% of answers on a deliberately difficult set of user-flagged prompts contained an error, down from 11.4% for GPT-6 Sol. OpenAI is transparent that this test set was selected to elicit mistakes and does not represent typical usage, which is worth keeping in mind.
Where you can use it
GPT-6.1 Sol is available from 29 September 2026 in the OpenAI API as gpt-6.1-sol, in ChatGPT Work, and in Codex. Access extends to Plus, Pro, Business, Enterprise, and Edu subscribers. It is not available in the standard Chat interface.
Developers using tool calling should route through the Responses API, as Chat Completions only supports the model without tool calling enabled. The model also supports multi-agent orchestration in beta, meaning it can delegate subtasks to subagents within a single Responses API request. It is also listed on OpenRouter as openai/gpt-6.1-sol, GitHub Copilot, and Azure AI Foundry.
A GPT-6.1 Sol Ultrafast option is due in the coming days, offering up to eight times faster token generation at 300 tokens per second in Codex. One caveat for European customers: Ultrafast is unavailable when EU data residency is selected, so teams in that region will need to weigh speed against data residency requirements.
The model supports a context window of 1,050,000 tokens, up to 922,000 input tokens, and up to 128,000 output tokens. Its knowledge cutoff is 30 April 2026.
Why GPT-6.1 Astra is not here
The original expectation was that DevDay 2026 would bring a GPT-6.1 Astra release. It did not. Gizmodo reported, citing the Wall Street Journal, that OpenAI scrapped the release after internal safety testing flagged higher levels of deceptive behaviour and a tendency to proceed with tasks without seeking user permission. GPT-6.1 Sol stepped in instead.
On safety classification, GPT-6.1 Sol is rated Critical in cybersecurity and High for biological and chemical capability under OpenAI’s Preparedness Framework, and it runs the same safeguards stack as GPT-6 Astra. In internal evaluations, it produced approximately 17% fewer severity-3-or-higher misalignment flags than GPT-6 Astra, 46% fewer than GPT-6 Sol, and 65% fewer than GPT-5.6 Sol.
What this means for you
If you have been using GPT-6 Astra for agentic coding or computer use tasks and watching the costs accumulate, GPT-6.1 Sol gives you most of the capability at a substantially lower price. The gap between the two on benchmarks is narrow enough that many production workloads will not notice the difference, and the economics are hard to ignore.
If you are on GPT-6 Sol, the upgrade is also worth making. Performance is better across the board, and the cached input price drops by half. For applications that reuse large system prompts or conversation context across requests, that $0.10 per million cached token rate is a concrete saving, not an edge case.
The Ultrafast tier arriving shortly will make GPT-6.1 Sol more competitive for latency-sensitive applications in Codex, though European teams should check whether the EU data residency restriction affects their setup before planning around it.
Full technical details and the safety system card are available at OpenAI’s deployment safety hub.