OpenAI Launches GPT-6.1 Sol, Bringing Near-Astra Agentic Performance to a Lower-Cost Tier

OpenAI launched GPT-6.1 Sol on September 29 as an upgrade to GPT-6 Sol, positioning it as a lower-cost model for coding, computer use, and professional work. The company says Sol nearly matches GPT-6 Astra on those tasks while charging about one-fifth of Astra’s standard input and output token prices. The release follows GPT-6 Sol by roughly 1 week, giving developers another model tier soon after the GPT-6 family expanded.

GPT-6.1 Sol is available immediately in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users, as well as through the OpenAI API under the name gpt-6.1-sol. The launch places a capable agentic model closer to routine production budgets, particularly for teams running large volumes of software, document, and desktop automation tasks.

GPT-6.1 Sol costs $2 per million input tokens, $10 per million output tokens, and $0.10 per million cached input tokens. OpenAI says cached input is priced 95% below standard input, while the standard rates are about 80% below GPT-6 Astra’s comparable pricing. The combination targets repeated workflows where long instructions, codebases, or document collections are sent to a model more than once.

The lower price matters because agentic systems often make many model calls while inspecting files, navigating applications, testing code, and revising results. Vellum’s analysis estimates GPT-6.1 Sol at about $0.38 per GDP.pdf task, compared with roughly $0.80 to $1.55 for Opus 5.5 and $1.95 for Astra. Those estimates frame Sol as a practical choice for document-heavy legal, compliance, and wealth-management pipelines, where accuracy and operating cost both shape deployment decisions.

GPT-6.1 Sol is designed for agentic coding, computer use, and multi-step professional tasks. OpenAI describes it as a substantial improvement over GPT-6 Sol, with gains in code writing and debugging, document understanding, and business workflows. Benchmark coverage highlighted by Vellum includes DeepSWE, OSWorld 2.0, GDP.pdf, and AutomationBench, spanning software engineering, desktop control, document analysis, and process execution.

That positioning gives Sol a role between lightweight automation models and OpenAI’s most expensive flagship tier. GPT-6 Sol was already aimed at complex coding and agentic work at scale, while GPT-6.1 Sol raises the performance ceiling without matching Astra’s price. Independent benchmark summaries from Artificial Analysis place the release within a broader GPT-6 lineup that trades intelligence, speed, and price across multiple models, making workload routing a central part of the product strategy.

OpenAI says GPT-6.1 Sol is more transparent about its limitations and more reliable at respecting user intent and safety constraints. The company published a deployment safety addendum covering the model’s data and training, safety evaluations, safe-completion behavior, and protections for users under 18. The addendum states that Sol uses the same types of data and training described for GPT-6 Astra.

The safety positioning carries added significance because Hindustan Times reported that OpenAI abandoned plans for a more advanced GPT-6.1 Astra version after testing found instruction-following and behavioral reliability problems. That account places Sol’s release alongside a more cautious model-development cycle, although the available launch coverage focuses mainly on performance and cost rather than independent validation of the safety claims. The system-card material provides the formal evaluation framework, while OpenAI’s product announcement supplies the commercial claims.

Developers gain a model aimed at sustained coding and computer-use workloads without paying flagship rates for every call. GPT-6.1 Sol’s API availability also supports direct integration into software agents, code-generation systems, document pipelines, and internal automation tools. Cached-input pricing is especially relevant to applications that repeatedly reference the same system instructions, repository context, or business records.

Enterprises can use Sol to route routine or high-volume tasks away from Astra while reserving the flagship model for work that demands its highest capability. That design can reduce inference spending, but it also puts pressure on teams to evaluate error rates, escalation rules, latency, and task-specific benchmark results rather than relying on a single intelligence label. The claimed near-Astra performance matters most when Sol maintains accuracy across long, multi-step workflows, where a cheap first pass can become expensive if human correction is frequent.

GPT-6.1 Sol intensifies competition around price-adjusted performance. OpenAI is selling a model that sits close to its flagship on selected professional tasks, while competitors continue to differentiate through coding strength, agent reliability, context handling, and inference economics. The launch also follows OpenAI’s introduction of GPT-6 Sol and Luna, indicating a rapid expansion of models tailored to distinct cost and capability bands.

The immediate test is adoption in production. Developers will measure whether Sol’s benchmark gains translate into fewer retries, safer computer actions, cleaner code, and lower total workflow cost. OpenAI’s release gives buyers a reason to reassess model routing now, while independent evaluations and real-world deployments will determine whether the near-Astra claim holds beyond the company’s selected tasks.