GPT-6.1 Sol: Near-Astra Performance at One-Fifth the Cost

Pricing

API pricing: standard input $2 per million tokens, output $10 per million tokens, cached input $0.10 per million tokens. This is 95% cheaper than standard input and 50% cheaper than GPT-6 Sol's cached input.

Performance Benchmarks

On DeepSWE v1.1, GPT-6.1 Sol high reasoning mode scores 75.2%, surpassing GPT-6 Sol max reasoning at 68.8%, with single-task cost reduced by approximately 76%.

On OSWorld 2.0 offline set, max reasoning scores 71.4% versus Astra's 73.5%, but single-task cost is roughly one-seventh of Astra's.

On AutomationBench, medium reasoning scores 31.7%, a 4.8 percentage-point improvement over GPT-6 Sol under the same settings.

Factual Accuracy and Safety

On difficult prompts, low reasoning mode reduces factual error rate from 11.4% to 7.7%, a roughly 32% reduction. Safety evaluations found no attempts to bypass safety reviews.

Availability and Upcoming Ultrafast Version

GPT-6.1 Sol is available today for Plus, Pro, Business, Enterprise, and Edu users on ChatGPT Work and Codex. The API model name is gpt-6.1-sol. OpenAI also announced an upcoming Ultrafast version with generation speeds up to 8 times faster than the standard version.

Note on GPT-6.1 Astra

Reportedly, OpenAI decided not to release GPT-6.1 Astra at the last moment due to safety concerns. The release of GPT-6.1 Sol provides a cost-effective alternative for developers, while Astra's release timeline remains uncertain.