OpenAI's GPT-6.1 Terra targets everyday computer work at half the API price of its predecessor
OpenAI's new mid-tier model focuses on computer use and visual reasoning, costs $1/$5 per million tokens, and is in ChatGPT Work and Codex.
The skim
- OpenAI released GPT-6.1 Terra, its mid-tier model, built for visual reasoning, computer use and clearer writing. It's in ChatGPT Work and Codex for paid plans, and in the API as
gpt-6.1-terra. - API pricing is $1 per million input tokens and $5 per million output tokens, which OpenAI says is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
- OpenAI's own benchmarks show big gains over the previous Terra, including a factual error rate that fell from 53.6% to 16.7% at low reasoning effort. Independent tests aren't out yet.
What OpenAI released
OpenAI's lineup runs, from cheapest to most capable, Luna, Terra, Sol and Astra. GPT-6.1 Terra replaces GPT-5.6 Terra in the middle slot. OpenAI calls it "a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing," aimed at everyday office work like drafting proposals, reading technical drawings, filling out forms and editing videos. Its example of high-volume work is reading incoming invoices and typing the details into accounting software.
The launch post says Terra is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. OpenAI's changelog also lists Business plans. It isn't in regular Chat: you pick it from the model list in ChatGPT Work or the desktop app, or run codex -m gpt-6.1-terra in the Codex CLI. Cached input costs $0.05 per million tokens.
The numbers OpenAI is claiming
- Computer use: On Agents' Last Exam, a benchmark of professional tasks, Terra at medium effort scores 15 percentage points higher than GPT-5.6 Terra at about a third of the cost, and matches GPT-6 Astra at a tenth of its price per token. OpenAI says it finishes tasks more than three times faster than Astra and Sol.
- Versus Anthropic: At maximum effort, OpenAI says Terra uses tokens more than three times as efficiently as Claude Opus 5.5 at a quarter of the price per token, for a cost per task about 9 times lower.
- Coding: It solves 50.3% of Terminal-Bench 4.0 tasks at max effort, up from GPT-5.6 Terra's 21.5%. OpenAI still recommends Sol or Astra for the hardest coding work.
- Accuracy: On OpenAI's internal test of error-inducing prompts, the error rate fell to 4.3% at max effort.
OpenAI also says Terra meets the "Critical" cybersecurity threshold in its Preparedness Framework and carries over the safeguards built for GPT-6 Astra.
Why it matters
Most AI work at companies is routine and runs in high volume, so the mid-priced model is often the one that matters most for budgets. Halving the price while improving computer use makes it cheaper to hand an agent an entire workflow, like clicking through a web form or a desktop app, instead of a single prompt. It also puts direct pressure on Anthropic, whose Opus 5.5 OpenAI chose as its comparison on the same day Anthropic filed to go public. The two have also been fighting on price for business customers.
All of these numbers are OpenAI's own, from its own tests. Treat them as claims until independent benchmarks and real-world use catch up.
Sources: OpenAI, OpenAI ChatGPT & Codex changelog, The Verge and 9to5Mac.
