OpenAI has released GPT-6.1 Sol, an updated AI model designed for agentic programming, computer control, and complex professional tasks. This version bridges the performance gap between the Sol line and the flagship GPT-6 Astra while offering significantly lower pricing.
The model is priced at $2 per million input tokens and $10 per million output tokens, which is five times cheaper than the Astra model. Additionally, cached input data costs $0.10 per million tokens, representing a 95% discount compared to standard input rates.
In performance benchmarks, GPT-6.1 Sol outperformed its predecessor, GPT-6 Sol, across multiple tests. It showed a 6.4 percentage point improvement in DeepSWE v1.1 and a 7 percentage point lead in OSWorld 2.0. It also surpassed Opus 5.5 in the AutomationBench and GDP.pdf evaluations.
The model demonstrates higher factual accuracy, with an error rate of 4.1% at high reasoning levels, compared to 4.5% for GPT-6 Sol. It also shows improved adherence to safety constraints and human intentions, with no recorded attempts to bypass security checks.
GPT-6.1 Sol is currently available to Plus, Pro, Business, Enterprise, and Edu users through ChatGPT Work and Codex. It is not yet accessible in the standard version of ChatGPT.
