OpenAI unveils GPT-6.1 Terra, a faster and cheaper AI model with less hallucinations

OpenAI has officially released GPT-6.1 Terra, the latest iteration in its GPT-6 series, marking a substantial leap forward in reducing hallucinations and improving computer task execution. The company reports dramatic improvements on internal benchmarks, with error rates on error-inducing prompts falling from 53.6% with the previous GPT-5.6 Terra model to just 16.7%, a reduction of nearly 37 percentage points. At maximum reasoning effort, errors dropped further to 4.3%, approaching the performance levels of OpenAI's higher-end [GPT-6 Astra](https://alternativeto.net/news/2026/9/openai-launches-gpt-6-astra-with-stronger-agentic-skills-and-tighter-cyber-safeguards/OpenAI launches GPT-6 Astra with stronger agentic skills and tighter cyber safeguards) and GPT-6.1 Sol models. The release emphasizes capabilities “from visual understanding to action”, positioning the model for real-world computational tasks.

The launch also includes reinforced safety measures and accessibility options. GPT-6.1 Terra meets OpenAI's “critical threshold” for cybersecurity risks under its Preparedness Framework and incorporates enhanced alignment improvements over earlier versions, including stronger resistance to prompt-injection attacks and better honesty about model limitations. The model is now available to Plus, Pro, Enterprise, and Edu users through ChatGPT Work and Codex interfaces, with developer access via the OpenAI API. Standard API pricing sits at $1 per million input tokens, $5 per million output tokens, with a notably discounted $0.05 rate for cached input tokens.