OpenAI releases GPT-6.1 Terra: GPT-6 Astra performance at one-tenth the token cost

OpenAI launched GPT-6.1 Terra on October 27, a medium-sized AI model designed to handle complex computer-based tasks at significantly lower cost than previous versions. The model excels in visual reasoning, desktop application automation, and natural language writing, targeting practical business workloads such as invoice processing, technical drawing interpretation, form completion, and video editing.

GPT-6.1 Terra achieves performance comparable to OpenAI's flagship GPT-6 Astra model on the Agents' Last Exam benchmark—a test of complex professional tasks like converting architectural drawings to 3D models—while charging approximately one-tenth the token price. Standard API pricing is $1 per million input tokens and $5 per million output tokens, representing a 50% reduction compared to GPT-5.6 Terra and GPT-6.1 Sol. The model completes tasks more than three times faster than both GPT-6 Astra and GPT-6.1 Sol.

Benchmark improvements are substantial across multiple dimensions. On Agents' Last Exam at medium reasoning effort, GPT-6.1 Terra scored 15 percentage points higher than GPT-5.6 Terra while costing roughly one-third as much per task. Error rates on OpenAI's internal factuality benchmark dropped from 53.6% to 16.7% at low reasoning effort, and to 4.3% at maximum effort—approaching GPT-6 Astra levels. On Terminal-Bench 4.0 (coding tasks), the model solved 50.3% of challenges at maximum effort, more than double GPT-5.6 Terra's 21.5%. On MMMU-Pro visual reasoning and OSWorld 2.1 desktop automation benchmarks, GPT-6.1 Terra outperformed its predecessor while remaining cost-competitive with higher-tier models.

The model is now available to ChatGPT Plus, Pro, Enterprise, and Edu users through ChatGPT Work and Codex, as well as to developers via the OpenAI API (identifier: gpt-6.1-terra). OpenAI retains the security safeguards from GPT-6 Astra, including resistance to prompt injection and cybersecurity preparedness at the Critical threshold. The company notes that GPT-6.1 Sol and GPT-6 Astra remain recommended for the most complex coding, scientific research, and deep reasoning tasks. OpenAI simultaneously announced three new AI-generated scientific results in mathematics, computer science, and physics, including proofs related to harmonic analysis, complexity theory, and quantum field theory.