OpenAI Prices GPT-6.1 Sol at $2 Input, $10 Output Per Million Tokens

OpenAI listed gpt-6.1-sol in its API on September 29, priced at $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. The model also shipped inside ChatGPT Work and Codex, but only for Plus, Pro, Business, Enterprise, and Edu accounts. Free and basic Chat users do not get it. That line matters more than the price itself. OpenAI is drawing a boundary around who touches its newest coding and computer-use model, and that boundary runs through paid seats, not through capability tiers inside a single product.

OpenAI says Sol reaches results close to Astra standard on selected coding, computer-use, and professional-work evaluations, at roughly one-fifth of Astra's token prices. That is OpenAI's claim, tied to evals it selected, not an independent benchmark. The announcement does not name which evals, how many tasks, or how Astra standard was configured for the comparison. It also does not say whether Sol's context window, latency, or tool-calling behavior matches Astra's. Those gaps matter for anyone deciding whether to swap a production workload from Astra to Sol based on cost alone.

For a manager running a coding or support workflow through the API, the arithmetic is straightforward to check and worth checking line by line. At $10 per million output tokens against Astra's roughly five-times-higher output price, a team generating heavy output, long code diffs, agent traces, transcripts, stands to cut spend fast if Sol's accuracy holds on its actual workload, not just on OpenAI's chosen evals. The risk sits in the word selected. A model can beat a benchmark average while missing on the edge cases a business depends on, and the only way to know is to run the team's own test set before repricing a contract or a budget line.

Workers who use Codex day to day get Sol automatically if their plan qualifies, with no separate purchase step described in the announcement. That is a win for individual productivity on paid plans. It is also a quiet shift in bargaining power: features tied to Work and Codex access increasingly separate from the free ChatGPT product, so a worker's capability now depends on which tier their employer buys, not on a personal account. Policy watchers should note that OpenAI did not publish usage caps, rate limits, or a rollback plan for Sol inside these products, details that matter if an employer standardizes workflows on a model with terms that can still change.

Enterprise and Edu buyers face a narrower question: does Sol replace Astra in existing contracts, or sit alongside it. OpenAI's announcement frames Sol as a cheaper near-match, not a stated replacement, and leaves pricing commitments, support terms, and data handling for Sol unaddressed in the text reviewed here. A business expanding Codex seats this quarter should ask OpenAI directly whether Sol pricing is locked for a term or subject to change, since the September 29 listing is a launch price, not a guaranteed rate.

The list price took effect September 29. Whether it holds through the next model cycle is the detail nobody has answered yet.