One signal

Update What changed

OpenAI says GPT-6.1 Terra is available now in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and through its API. That turns the earlier announcement into a live option for developers and subscribers.

OpenAI lists standard API prices of $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens. The company says those rates are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol. It gives invoice reading and data entry as examples of high-volume tasks the lower cost could make practical.

The company also reports that Terra scored 15 percentage points above GPT-5.6 Terra on its Agents’ Last Exam computer-use benchmark at medium reasoning effort, at about one-third the cost. OpenAI says Terra completed tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol.

Sources and evidence

• Introducing GPT-6.1 Terra - OpenAI: OpenAI says GPT-6.1 Terra is available in ChatGPT Work, Codex and the API, with standard API prices of $1 per million input tokens, $0.05 per million cached input tokens and $5 per million output tokens; it also reports lower cost and faster performance on specified benchmarks.

Independent WittyWires Watcher; not an official account or feed.

Update What changed

OpenAI says GPT-6.1 Terra solved 50.3% of tasks on Terminal-Bench 4.0 at maximum reasoning effort, compared with 21.5% for GPT-5.6 Terra. The benchmark tests coding agents on command-line work, including fixing a webpage whose layout shifts as it loads.

The company also reports a lower error rate on its internal benchmark of prompts designed to induce mistakes: 16.7% at low reasoning effort, down from 53.6% for GPT-5.6 Terra. At maximum effort, it says Terra’s rate was 4.3%. These are OpenAI’s own benchmark results, not independent tests.

OpenAI says GPT-6.1 Sol and GPT-6 Astra remain better choices for its most complex coding tasks. That is a useful boundary on the coding claim: Terra’s reported gains do not make it the top choice for every job.

Sources and evidence

• GPT-6.1 Terra: OpenAI says GPT-6.1 Terra scored 50.3% on Terminal-Bench 4.0 at maximum reasoning effort and lowered its internal low-effort error rate to 16.7%, compared with 53.6% for GPT-5.6 Terra.

Independent WittyWires Watcher; not an official account or feed.

Update What changed

Simon Willison’s assessment gives GPT-6.1 Terra a more specific job description: a mid-range option for visual reasoning, computer use and technical communication, rather than a general-purpose upgrade for every task. He points to drafting proposals, interpreting technical drawings, filling in forms and editing videos as examples of the work it may suit.

Willison also relays OpenAI employee Ted Sanders’s comparison: Sanders said Terra was probably better than Astra and Sol at technical communication, similar on vision and computer use but faster, and weaker on most other things.

For developers choosing a model, the useful distinction is workload, not just the price tag: Terra may be the sensible pick for repetitive computer-based tasks or clearer technical writing, while more complex work may favour another model.

Sources and evidence

• GPT-6.1 Terra: Simon Willison describes GPT-6.1 Terra as a mid-range model suited to visual reasoning, computer use and technical communication, and relays Ted Sanders’s view that it compares favourably on technical communication but is weaker on most other tasks.

Independent WittyWires Watcher; not an official account or feed.