OpenAI Unleashes GPT-6.1 Sol: Near-Astra Intelligence at a Fraction of the Cost for Developers

OpenAI's GPT-6.1 Sol is positioned as a mid-tier reasoning model within the GPT-6 series, slotting between the most capable GPT-6 Astra and the more efficient, lower-cost GPT-6 Luna. The core appeal of GPT-6.1 Sol lies in its ability to deliver 'near-Astra intelligence' for critical developer tasks such as agentic coding, computer use, and professional workflows, all while significantly cutting down on operational costs.

According to OpenAI, GPT-6.1 Sol matches GPT-6 Astra's performance on complex software engineering tasks as evaluated by DeepSWE v1.1, but at roughly one-fifth of the cost. This remarkable cost-performance ratio is a direct response to the growing demand for affordable enterprise-grade AI, making sophisticated AI capabilities accessible to small startups and individual developers who might have found previous flagship models too expensive for routine work.

Furthermore, the model boasts a substantial 1,050,000-token context window, allowing developers to feed entire codebases, extensive documentation, or lengthy research materials into the model within a single request. This massive context window is complemented by a maximum output of 128,000 tokens, enabling GPT-6.1 Sol to generate comprehensive and detailed responses, which is crucial for complex software development tasks like code generation, debugging, and refactoring.

A significant improvement for developers building AI agents is the revamped pricing for cached input. OpenAI has reduced the cached input cost to $0.10 per million tokens, a 95% discount off the standard input price and half of GPT-6 Sol's cached input price. This is particularly impactful for agentic workflows where the AI agent frequently re-reads the same context, such as a codebase or project documentation, leading to substantial savings over long-running tasks.

GPT-6.1 Sol is engineered for demanding work across code, documents, and applications. Its capabilities extend beyond simple text generation, making it a robust tool for building advanced AI agents. The model supports various reasoning efforts—low, medium (default), high, xhigh, and max—allowing developers to fine-tune its cognitive load based on the complexity and criticality of the task.

For coding, GPT-6.1 Sol excels in software engineering tasks across large and complex codebases, including code generation, debugging, refactoring, issue tracing, and repository-level changes. On the DeepSWE v1.1 benchmark, which evaluates complex software engineering tasks in real codebases, GPT-6.1 Sol matches GPT-6 Astra's performance at a fraction of the cost. This makes it an ideal choice for developers looking to automate significant portions of their development lifecycle or create intelligent coding assistants.

Beyond coding, GPT-6.1 Sol demonstrates substantial progress in tasks requiring interaction with computer applications. On OSWorld 2.0's offline set, which evaluates agents on demanding computer-use workflows, it comes within 2.1 percentage points of Astra's score while costing approximately one-seventh as much per task. This indicates its strong potential for automating multi-step business workflows, document understanding, and general computer use.

The model's improved factuality at low reasoning effort is another notable feature, reducing the share of responses containing factual errors by approximately 32% compared to GPT-6 Sol. This enhanced reliability is crucial for applications where accuracy is paramount, such as scientific research and critical business processes.

Developers can access GPT-6.1 Sol through the OpenAI API using the model ID gpt-6.1-sol. The standard API pricing is set at $2.00 per million input tokens and $10.00 per million output tokens, which aligns with GPT-6 Sol's pricing but offers significantly better performance. The highly discounted cached input rate of $0.10 per million tokens is a key differentiator for agent-based applications.

It's important to note that tool calling with GPT-6.1 Sol requires the Responses API, and the older Chat Completions API only works without tools for this model. Additionally, the none and minimal reasoning efforts are no longer supported, with low being the lowest available setting. Developers currently using GPT-6 Sol for function calling at none effort will need to adapt their integrations.

OpenAI also announced an 'Ultrafast' tier for GPT-6.1 Sol, promising up to 8x faster token generation in Codex, which will be available in the coming days. This high-speed option caters to applications requiring extremely low latency. GPT-6.1 Sol is available to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, but not yet in the main ChatGPT chat interface.

For teams evaluating production AI, the key takeaway is to test GPT-6.1 Sol against their specific tasks. While GPT-6 Astra still leads on the hardest cyber, biology, and science evaluations, GPT-6.1 Sol offers a compelling trade-off for most agentic coding, computer use, and business workflows, making it a strong candidate for the 'daily workhorse' model.