OpenAI announces GPT-6.1 Sol: a powerful and economical coding agent
OpenAI has introduced the updated GPT-6.1 Sol model, which significantly improves agent programming and code handling. The model is designed for complex tasks such as deep refactoring and the creation of long-lived autonomous agents.
GPT-6.1 Sol offers an impressive combination of performance and cost-effectiveness. While Astra is recommended for tasks where quality is critical, GPT-6.1 Sol is ideal for frequently running complex workflows, as it provides performance close to Astra at significantly lower costs.
In addition to increasing efficiency, the model also lowers data processing costs: cached input costs 95% less than the standard price, making it ideal for resource-intensive but repetitive tasks. In DeepSWE v1.1 tests, GPT-6.1 Sol outperformed GPT-6 Sol by 76% at approximately 3.5 times lower cost per task.
Why it matters
- —The model significantly increases efficiency in coding and working with codebases, which is critical for developers.
- —The introduction of ultra-cheap cached input allows running complex, resource-intensive workflows much more frequently.
- —GPT-6.1 Sol offers an optimal balance between high performance and low cost, making it ideal for commercial use.
Key facts
- In DeepSWE v1.1, GPT-6.1 Sol achieved a score of 75.2% under high effort, surpassing GPT-6 Sol (68.8%) at lower costs.
- The cost of cached input is only $0.10 per million tokens, which is 95% cheaper than standard input.
- The model reduces the number of factual errors in responses by approximately 32% compared to GPT-6 Sol.
- On AutomationBench, GPT-6.1 Sol scored 31.7% at an average effort level.
The full text is in the original source. Here we provide a brief summary and key facts.