OpenAI launches GPT-6.1 Terra with cheaper computer use and coding

What happened?

OpenAI released GPT-6.1 Terra on October 26, positioning it around visual reasoning, computer use and everyday writing. The company says it is available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users; its API changelog confirms the release date.

Source: OpenAI ↗ · Source: OpenAI Developers ↗

What does GPT-6.1 Terra cost?

Standard API pricing is $1 per million input tokens, $0.05 for cached input and $5 for output. The input and output rates are half GPT-6.1 Sol’s listed standard rates of $2 and $10. These are usage prices, separate from a ChatGPT subscription.

The headline rates do not cover every configuration. The model documentation lists higher charges for prompts exceeding 272,000 input tokens, separate cache-write prices and a premium for Fast mode. Comparing completed tasks requires accounting for the full request and any tools used.

Source: OpenAI Developers ↗ · Source: OpenAI Developers ↗

What can developers actually use?

The API model identifier is gpt-6.1-terra. Its documentation lists a 1,050,000-token context window, up to 128,000 output tokens, text and image input, and text output. Native audio and video input are not supported.

Tool calling requires the Responses API. Listed tools include computer use, web and file search, code interpreter, hosted shell and MCP. Reasoning effort runs from none through max, with medium the default. These documented interfaces describe what developers can connect, rather than guaranteeing success on a particular workflow.

Source: OpenAI Developers ↗

The benchmark gains need their context

OpenAI reports 50.3% on Terminal-Bench 4.0 at maximum reasoning effort, compared with GPT-5.6 Terra’s published 21.5%. On Agents’ Last Exam at medium effort, it reports a 15-percentage-point gain over GPT-5.6 Terra at about one-third the cost. Those are company-reported comparisons, not LessCarbs tests.

The launch also reports a low-effort factual-error reduction from 53.6% to 16.7% on an internal evaluation. The system card explains that its factuality evaluation deliberately uses difficult, user-flagged examples. Its absolute error rates should not be read as the frequency of hallucinations in ordinary production use.

Source: OpenAI ↗ · Source: OpenAI Deployment Safety Hub ↗

A lower price does not remove the safety questions

OpenAI says it is treating Terra as Critical capability in cybersecurity under its Preparedness Framework and applying the safeguards stack used for GPT-6.1 Sol and GPT-6 Astra. That is a capability assessment, rather than a statement that every deployment is unsafe.

The system card also cautions that some alignment evaluations may not directly represent real-world behavior. Its deployment simulation records higher rates of several kinds of problematic behavior for Terra than for Sol and Astra. The evidence supports comparing specific tasks and failure modes, rather than treating alignment as a single ranking.

Source: OpenAI Deployment Safety Hub ↗

Why does it matter?

Our analysis: the useful question is how cheaply a model can finish work that someone can verify. A lower token rate creates room to try more routine tasks, but savings can disappear if a workflow repeatedly fails, needs extra attempts or leaves a person repairing the result.

For a team considering Terra, a meaningful trial would use representative documents and screens, define what counts as a correct result, and measure completion time alongside review effort. Check the saved spreadsheet, changed code or submitted record itself. A confident completion message is not the deliverable.

The release gives developers another price-and-capability option. Whether it improves a real workflow will depend on the work assigned, the tools connected and the evidence used to judge completion. The launch benchmarks are a starting point for that comparison.