OpenAI unveiled GPT-6.1 Sol at its DevDay on September 29, positioning it for coding agents, computer use, and professional office tasks. Standard API pricing is $2 per million input tokens, $0.10 for cached input, and $10 per million output tokens — unchanged from GPT-6 Sol. Against the flagship GPT-6 Astra's $10, $0.20, and $50, both input and output pricing land at one-fifth.
The release cadence has been rapid. GPT-6 Astra shipped 26 days earlier, and GPT-6 Sol had been out for just seven days before being replaced by the 6.1 version. OpenAI says the new model comes "close to Astra" on coding and computer-use tasks.
Benchmarks near the flagship, with some gaps remaining
In OpenAI's own results, GPT-6.1 Sol scored 75.2% on DeepSWE v1.1 at the higher reasoning setting, matching Astra while costing roughly a fifth as much. On OSWorld 2.0 at the highest reasoning tier, it scored 71.4%, 2.1 percentage points behind Astra, at about one-seventh the cost per task. On Terminal-Bench Science 0.1, which tests command-line research workflows, it averaged $5.47 per task, versus $23.21 for Claude Opus 5.5 and $23.80 for Astra.
Independent testing from Artificial Analysis reached similar conclusions: its intelligence index is 4 points higher than GPT-6 Sol and 1 point lower than GPT-6 Astra; on a knowledge-accuracy test, the hallucination rate fell from 60% to 54%. The trade-off is verbosity — completing the same task takes 10% to 30% more output tokens. A jump in the cached-read discount from 90% to 95% offsets part of that cost.
Not every category is this close. In OpenAI's own figures, Sol scored 21.5% on the internal ExploitBench versus Astra's 31.5%, and 47.96% versus 63.46% on TroubleshootingBench. For scenarios requiring long chains of debugging, the flagship still holds a clear edge.
Access and a few limits
GPT-6.1 Sol is now available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users; free-tier and Go users don't have access yet, and it hasn't reached the regular chat interface. Developers can call it as gpt-6.1-sol, with a context window of roughly 1.05 million tokens and a maximum output of 128,000 tokens per call.
A few details worth noting:
- For prompts beyond 272,000 tokens, input is billed at 2x and output at 1.5x.
- The none and minimal reasoning-effort tiers are not supported.
- Tool calls aren't available through the Chat Completions endpoint; teams need to switch to the Responses API.
The second point matters most for teams building real-time interactions. The old trick of turning off reasoning to cut latency no longer works on 6.1 Sol.
Where developers in China can access it
OpenAI's official API has been unavailable in mainland China since July 2024, and that hasn't changed. The third-party channels on this launch's list are more relevant to developers there: OpenRouter and Vercel AI Gateway added it at the same time, and it's also available through GitHub Copilot's Pro+, Max, Business, and Enterprise tiers.
For teams in China, the more direct comparison is domestic model pricing. GLM-5.3's published API price is $4.40 per million output tokens — less than half of GPT-6.1 Sol's. By pushing flagship-level coding capability down to the $10 tier, Sol has narrowed the price gap that domestic models had opened up in coding use cases; whether they respond with price cuts of their own remains to be seen.
Another variable sits with OpenAI itself. The company pulled its originally planned October release of GPT-6.1 Astra the day before, citing safety standards not yet met, with no timeline set for the flagship line's next update. Until then, 6.1 Sol is likely to be the model OpenAI pushes hardest to developers.
Sources: OpenAI's official release page and DevDay recap, Artificial Analysis independent benchmarks, CocoLoop, benchmark roundups from DataCamp and Vellum; per-tier API pricing, context length, and long-prompt billing multipliers follow OpenAI's published figures.