
gpt-6.1-terra
API Overview
GPT-6.1 Terra is OpenAI’s mid-range flagship model in the GPT-6.1 generation, designed for high-concurrency everyday agentic tasks, visual understanding of screens/UI images, and cost-effective code generation. While maintaining low API call costs, Terra delivers visual reasoning capabilities that approach those of the flagship Astra, along with multi-step screen automation efficiency. It is the industrial-grade, cost-effective choice for enterprises deploying desktop/web automation agents at scale, as well as for screen/form parsing and large-scale everyday writing.
───────────────────────────────────────────────────────────────────
Core Capabilities
State-of-the-art visual reasoning and screen interaction: Achieves 85.5% on MMMU-Pro (Vision), 1.2 percentage points lower than Astra; in the Agents' Last Exam v1 automation task test, it boasts a high win rate and reduces single-task execution time by more than threefold.
Low hallucination and robust adversarial error prevention: In OpenAI’s official highly adversarial prompt tests, the error rate under Max inference mode is significantly reduced to 4.3%.
Highly competitive compute costs: The standard API rate is 50% of Sol’s, and the Prompt Caching read rate is as low as $0.05 per 1 million tokens, making it ideal for high-frequency agent loops that frequently load system prompts.
1.05M ultra-long context and 128K output: Supports an ultra-large context capacity of 1,050,000 tokens, with a maximum single-generation output of up to 128,000 tokens.
Playground
Log in to explore more features! Click to Log In