GPT-6.1 Now Available via API at $0.10 per Million in Cache

OpenAI has made GPT-6.1 Sol available for professional offers and through its API, with significantly reduced pricing, particularly for cached input. The model is reported to be approaching Astra on several evaluations while lowering the cost per task. OpenAI also highlights improvements in factual accuracy and alignment, and is preparing an Ultrafast variant.

Immediate Access and Pricing Structure Announced by OpenAI

GPT-6.1 Sol is available today for Plus, Pro, Business, Enterprise, and Edu subscribers within ChatGPT Work and Codex. It is not yet offered in Chat. Developers can access it via the OpenAI API under the identifier gpt-6.1-sol. The standard rates communicated are $2 per million tokens of input, $10 per million tokens of output, and $0.10 per million for cached input. OpenAI also plans to launch GPT-6.1 Sol Ultrafast in the coming days, which will offer token generation up to eight times faster than the standard speed in Codex.

Coding and Application Usage: Results and Cost Disparities

OpenAI presents GPT-6.1 Sol as approaching the intelligence of GPT-6 Astra for agentic coding, computational use, and professional work, at significantly lower costs. On DeepSWE v1.1, GPT-6.1 Sol matches Astra for about one-fifth of the cost and surpasses the best score of GPT-6 Sol by 6.4 percentage points, with less reasoning effort. On OSWorld 2.0 offline, GPT-6.1 Sol outperforms GPT-6 Sol by seven percentage points at maximum effort for less than half the cost, and is 2.1 percentage points away from Astra's score for about one-seventh of the cost per task.

Professional Tasks and Science: Comparisons with Opus 5.5 and Astra

For complex professional tasks, OpenAI reports that in the GDP.pdf test, GPT-6.1 Sol exceeds Opus 5.5 in terms of score while maintaining a cost per task of less than half in the evaluated reasoning parameters, achieving a performance level close to Astra for about one-fifth of the cost. On Terminal-Bench Science 0.1, GPT-6.1 Sol more than doubles the score of GPT-6 Sol at maximum effort, with an average cost of $5.47 per task, compared to $23.21 for Opus 5.5 and $23.80 for Astra, which is over 75% less than both models.

Factual Accuracy and Alignment: Reported Reductions in Errors and Failures

OpenAI emphasizes an improvement in factual accuracy with low reasoning effort, with the share of incorrect responses dropping from 11.4% to 7.7%, representing about a 32% reduction. In the evaluated reasoning parameters, the error rate gap between GPT-6.1 Sol and Astra remains at 1.9 percentage points, with a cost per task less than one-fifth of that of Astra. On the safety front, OpenAI reports advancements in alignment compared to GPT-6 Sol, increased transparency regarding limitations, and a more consistent adherence to user intentions and safety constraints, with lower failure rates for complex situations such as detecting faulty search tools, enforcing explicit restrictions, and avoiding prohibited outcomes.

Cost Structure and Caching: Reductions Communicated by OpenAI

OpenAI offers GPT-6.1 Sol at a rate equivalent to about one-fifth of Astra's standard input and output prices. The rate for cached input is set at $0.10 per million tokens, representing a 95% reduction from standard input prices and a 50% reduction compared to the rate of GPT-6 Sol. This pricing policy aims to provide developers with greater flexibility to design agents capable of reusing context between queries.