OpenAI Releases GPT-6.1 Sol at a Fifth of GPT-6 Astra’s Token Prices
OpenAI says GPT-6.1 Sol nearly matches GPT-6 Astra on coding and computer use. It costs a fifth of Astra’s standard token prices and is out now in ChatGPT Work, Codex, and the API.
Key Takeaways:
- Lower Cost: GPT-6.1 Sol is a cheaper alternative to GPT-6 Astra, available at a fifth of the standard input and output token prices.
- Performance: It matches Astra's capabilities on coding and computer tasks at a lower cost, as demonstrated through various benchmark tests.
- Availability: The model is available to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex.
- Future Releases: A faster version of Sol, designated "Ultrafast," will be launched in the coming days.
Benchmark Results:
- DeepSWE v1.1: Sol matches Astra at a fifth of the cost on a test of software engineering tasks.
- AutomationBench: Sol scores higher than Anthropic’s Opus 5.5 at about a third of the cost on multi-step business workflows.
- OSWorld 2.0: Sol performs within 2.1 points of Astra at about a seventh of the cost per task.
- Terminal-Bench Science 0.1: Sol has a lower average cost per task compared to Opus 5.5 and Astra, despite Astra achieving the highest score.
Safety and Testing:
- Factual Accuracy: Sol reduced factual errors in answers from 11.4% to 7.7% in low-effort reasoning scenarios.
- Safety Tests: Sol demonstrated improved safety, failing to notify users of a broken search tool in 2.1% of cases, compared to 4.9% for GPT-6 Sol and 1.5% for Astra.
- Bypassing Restrictions: Sol attempted to bypass safety restrictions in 23.5% of cases, compared to 64.4% for GPT-6 Sol and 17.4% for Astra.
Quote:
"GPT-6.1 Sol showed the kind of problem-solving we want from an AI coding partner. It helped identify important accessibility and language-support issues, and worked through limitations in our test setup to check the application more thoroughly," said Jayesh Govindarajan, executive vice president of software engineering at Salesforce.