OpenAI Launches GPT-6.1 Terra, Halving GPT-6.1 Sol’s API Price for Agent Work
MarkTechPost Michal Sutter ● Covered by 6 sources
OpenAI’s GPT-6.1 Terra is out, with cheaper API pricing for visual and computer-use work. It’s aimed at high-volume agent tasks, but coding still trails the best models.
Based on reporting by MarkTechPost, Michal Sutter — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI has launched GPT-6.1 Terra, a closed-weight model built for visual reasoning, computer use and writing. It replaces GPT-5.6 Terra and is available in the API as gpt-6.1-terra. The headline here is price: Terra is set at $1 per million input tokens and $5 per million output tokens, with batch and Flex pricing cut in half.
Terra is not a tiny specialist. It takes text and images, returns text, and ships with a 1,050,000-token context window plus up to 128,000 output tokens. The model runs only on OpenAI-hosted products — API, ChatGPT Work and Codex — and its knowledge cutoff is June 1, 2026. Reasoning effort ranges from none to max, with medium as the default. Tool use goes through the Responses API, and OpenAI lists computer use, hosted shell, code interpreter, web search, file search and MCP among the supported tools.
The pitch is volume work: matching product photos to catalog entries, reading technical drawings, and handling screens, forms and documents at scale. On OpenAI’s Agents’ Last Exam v1, Terra scores 57.6% at medium effort at $0.29 per task, matching GPT-6 Astra’s score while costing far less. At max effort, it reaches 60.0% at $1.01 per task. OpenAI says it also finishes tasks more than 3 times faster than GPT-6 Astra and GPT-6.1 Sol.
Vision looks strong too. Terra hits 85.5% on MMMU-Pro (Vision) at max effort, only 1.2 points behind GPT-6 Astra, and the run costs $0.012 per task versus Astra’s $0.065. On OpenAI’s internal error-prone prompt set, the low-effort error rate drops from 53.6% to 16.7%, and max-effort errors fall to 4.3%. But the coding story is less glossy. On Terminal-Bench 4.0, Terra gets 50.3% at max effort, well behind Claude Sonnet 5.5’s 70.6%. OpenAI says Sol or Astra are still the picks for the hardest coding jobs.
My take — AI-written commentary, not fact-checked reporting
This is the classic OpenAI move: make the agent model cheap enough that companies can’t ignore it, then let the pricier siblings handle the fancy bragging rights. Terra looks like the boring winner here, which is usually where the real money is. The model race is getting less about raw scores and more about who can do useful work without burning cash like a startup with a fresh funding round.
Read more about this at: MarkTechPost