OpenAI
APIs and tools for building AI products
5.0•882 reviews•38K followers
APIs and tools for building AI products
5.0•882 reviews•38K followers
The most powerful platform for building AI products.
Build and scale AI experiences powered by industry-leading models and tools.
This is the 53rd launch from OpenAI. View more
GPT-6.1 Terra
Launched this week
GPT-6.1 Terra is OpenAI's fast, efficient model for visual reasoning, precise computer use and natural writing. It reads what's on screen, takes action across apps, and handles workday tasks like filling out forms, interpreting technical drawings and drafting proposals. OpenAI says it matches GPT-6 Astra on Agents' Last Exam at one tenth the price per token and finishes tasks over 3x faster. Available now in ChatGPT Work and Codex, and in the API as gpt-6.1-terra.
Payment Required
Launch tags:Artificial Intelligence
Launch Team
Gemini 2.5 Pro
Claude Code
Hey PH community 👋
Excited to hunt GPT-6.1 Terra today.
OpenAI just shipped a new Terra model built for fast, capable computer use, and they say it matches GPT-6 Astra on Agents' Last Exam at one tenth the price per token while finishing tasks over three times faster. It reads what's on screen, takes action across apps and writes more naturally than GPT-5.6 Terra. OpenAI posted a short look at it on X here: https://x.com/OpenAI/status/2114780711664398349
**Computer use** is the part I'm most excited about. A model that can look at a screen, understand it and click through real apps at this speed and price makes it practical to hand off the repetitive work that fills a normal day, like entering invoices into accounting software or filling out forms.
**Visual reasoning** is the other one I'd test first. OpenAI says it nearly matches their top models on MMMU Pro, which matters for things like reading technical drawings, matching product photos to a catalog or reviewing video shot by shot.
**Factuality** also took a real jump. OpenAI says the error rate on their hardest internal prompts fell from 53.6% to 16.7% at low reasoning effort, which is exactly what you want from a model you plan to run at high volume.
For builders, it's already live in the API as gpt-6.1-terra, and API prices are 50% lower than GPT-5.6 Terra.
You can try it in ChatGPT Work and Codex right now: https://chatgpt.com
Give it a spin and let the team know what you create with GPT-6.1 Terra 🚀
@thisiskp_ Talk about perfect timing1 My Codex Pro200 allowance is at 19% w/ 6 days left before reset 😅. Will be using Terra 6.1 for the rest of the week I suppose. 😸
@alchemist_studio do it! Great timing
@thisiskp_ So far so good. its blazing fast, so great when you need speed, but needs an extra push here and there to keep going continiously on longer tasks. and more complex tasks, it probably uses more tokens and works longer than the big brother sol.
For computer use, the part I'd want to test first is recovery: when Terra misreads a screen or clicks the wrong control, does it notice and backtrack on its own, or does it keep going? With forms and technical drawings, a small grounding mistake can cascade, so I'm curious whether the API exposes the action trace or screenshots it used, so a calling agent can audit and retry a step instead of restarting the whole task.