OpenAI releases GPT-6.1 Sol with performance near GPT-6 Astra at one-fifth the cost

OpenAI has launched GPT-6.1 Sol, an upgraded model derived from GPT-6 Sol that achieves performance comparable to GPT-6 Astra in agentic coding and professional tasks. The new model is priced at one-fifth the cost of its predecessor, with cached input tokens priced at 0.10 USD per million tokens.

The pricing structure for GPT-6.1 Sol is set at 2 USD per million input tokens and 10 USD per million output tokens. This compares to 10 USD and 50 USD respectively for GPT-6 Astra. The cached input price of 0.10 USD per million tokens is 95 percent lower than the model's own standard input price and 50 percent lower than the cached input price of GPT-6 Sol. OpenAI states that this pricing allows developers more room to build and run AI agents that require repeated context across requests.

On the DeepSWE v1.1 benchmark, which measures complex software engineering tasks on real codebases, GPT-6.1 Sol scores on par with GPT-6 Astra at a cost of roughly one-fifth. It also scores 6.4 percentage points higher than the best score of GPT-6 Sol, using lower reasoning effort and cost. In scientific research tasks on Terminal-Bench Science 0.1, GPT-6.1 Sol scores more than double the score of GPT-6 Sol at maximum reasoning effort. The average cost per task is 5.47 USD, compared to 23.21 USD for Opus 5.5 and 23.80 USD for GPT-6 Astra, which is more than 75 percent lower than both models. The model also performs better in computer use tasks and answering professional questions from complex PDF documents.

Despite these improvements, GPT-6 Astra still achieves the highest score among the tested models at 68.1 percent. OpenAI recommends using GPT-6 Astra for the most difficult scientific research tasks. Regarding safety, OpenAI indicates that GPT-6.1 Sol performs better than GPT-6 Sol in alignment evaluations and moves closer to GPT-6 Astra. Full details are available in the GPT-6.1 Sol system card addendum.

GPT-6.1 Sol is available today for Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, but is not yet supported in Chat. Developers can access the model through the OpenAI API using the model name gpt-6.1-sol. In the coming days, OpenAI will release GPT-6.1 Sol Ultrafast, which generates tokens at a speed up to 8 times faster than the maximum standard speed in Codex.