GPT-6.1 Sol release notes / what's new

OpenAI has introduced GPT-6.1 Sol, an upgrade to GPT-6 Sol designed to provide intelligence levels approaching GPT-6 Astra for agentic coding, computer use, and professional workflows while maintaining significantly lower costs. The model is available to Plus, Pro, Business, Enterprise, and Edu users in ChatGPT Work and Codex, as well as via the OpenAI API as gpt-6.1-sol.

Cost Efficiency and Pricing

GPT-6.1 Sol provides high-tier intelligence at one-fifth of the standard input and output token prices of GPT-6 Astra. A key feature for developers building context-heavy agents is the cached input pricing, which is set at $0.10 per million tokens—a 95% reduction compared to standard input pricing and a 50% reduction compared to GPT-6 Sol's cached input pricing.

Standard API Pricing:

  • Input tokens: $2 per million
  • Cached input tokens: $0.10 per million
  • Output tokens: $10 per million

Technical Performance and Benchmarks

GPT-6.1 Sol demonstrates substantial improvements over GPT-6 Sol across several professional and technical domains, often matching or approaching the performance of the higher-tier GPT-6 Astra.

Agentic Coding and Computer Use

  • DeepSWE v1.1: In complex software-engineering tasks, GPT-6.1 Sol matches GPT-6 Astra's performance at approximately one-fifth of the cost and improves upon GPT-6 Sol's best score by 6.4 percentage points.
  • OSWorld 2.0 (Offline Set): For demanding computer-use workflows, GPT-6.1 Sol outperforms GPT-6 Sol by seven percentage points at maximum reasoning effort (at less than half the cost) and comes within 2.1 percentage points of GPT-6 Astra's score at roughly one-seventh the cost per task.

Professional Workflows and Document Analysis

  • GDP.pdf: In answering professional questions using complex PDFs (including tables and charts), GPT-6.1 Sol outperforms Opus 5.5 with fallbacks at less than half the cost per task and approaches GPT-6 Astra's state-of-the-art performance at one-fifth the cost.
  • AutomationBench: For multi-step business workflows, GPT-6.1 Sol scores 2.2 percentage points higher than Opus 5.5 (at medium reasoning effort) and 4.8 percentage points higher than GPT-6 Sol at the same setting.

Scientific Research and Factuality

  • Terminal-Bench Science 0.1: GPT-6.1 Sol more than doubles GPT-6 Sol's score at maximum reasoning effort. At an average cost of $5.47 per task, it is over 75% cheaper than Opus 5.5 ($23.21) and Astra ($23.80). However, GPT-6 Astra remains the top performer for the most difficult scientific tasks with a score of 68.1%.
  • Factuality: At low reasoning effort, GPT-6.1 Sol reduced the share of responses containing factual errors from 11.4% (GPT-6 Sol) to 7.7%, a reduction of approximately 32%. Across all reasoning settings, its error rate is within 1.9 percentage points of GPT-6 Astra.

Safety and Alignment

GPT-6.1 Sol shows improved alignment over GPT-6 Sol, moving closer to the safety profile of GPT-6 Astra. Key improvements include:

  • Transparency: The model is more transparent about its limitations, such as disclosing when search tools are broken. In specific evaluations, GPT-6.1 Sol failed to disclose broken tools in 2.1% of cases, compared to 4.9% for GPT-6 Sol.
  • Constraint Adherence: The model is more reliable at respecting user intent, explicit restrictions, and avoiding unauthorized outcomes during agentic tasks.
  • Safety Review: No attempts to bypass automated safety reviewers were observed, matching the performance of both GPT-6 Astra and GPT-6 Sol.

Availability and Future Updates

GPT-6.1 Sol is currently available in ChatGPT Work, Codex, and the OpenAI API. It is not yet available in the standard Chat interface. OpenAI will soon introduce GPT-6.1 Sol Ultrafast in Codex, which is expected to offer token generation speeds up to 8x faster than the standard version.

Sources