OpenAI released GPT-6.1 Terra at half the API price of GPT-5.6 Terra

A model priced for a hero demo is a different thing from one priced to run on every ticket. The gap is set by the standard API price per million tokens. OpenAI has moved that number.

OpenAI released GPT-6.1 Terra, aimed at computer use and everyday coding. The company says the standard API prices are 50 percent lower than GPT-5.6 Terra and GPT-6.1 Sol. The model is now in ChatGPT Work, Codex and the OpenAI API. GitHub says it is generally available in GitHub Copilot.

What OpenAI says the model does

OpenAI reports the model is faster. The company says Terra finishes tasks over three times faster than GPT-6 Astra and GPT-6.1 Sol. On Agents' Last Exam at medium reasoning effort, OpenAI reports Terra scores 15 percentage points higher than GPT-5.6 Terra. OpenAI says that is at about one-third of the cost per task. The company also says Terra matches GPT-6 Astra on this benchmark. OpenAI puts the price gap at one-tenth of GPT-6 Astra's per-token cost.

OpenAI says its internal error-rate benchmark fell at low reasoning effort. The rate went from 53.6 percent with GPT-5.6 Terra to 16.7 percent with GPT-6.1 Terra. At maximum reasoning effort, OpenAI says the error rate is 4.3 percent. On Terminal-Bench 4.0 at maximum reasoning effort, OpenAI reports Terra solves 50.3 percent of tasks. OpenAI compares that against GPT-5.6 Terra's published score of 21.5 percent. Every figure in this section is OpenAI's own benchmark. No independent test has been published.

Pricing and where to call it

OpenAI lists the standard API pricing at 1 dollar per million input tokens. Cached input tokens are 5 cents per million. Output tokens are 5 dollars per million. The developer identifier is gpt-6.1-terra. The model is available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and in Codex.

GitHub Copilot has it

GitHub said in a changelog post that GPT-6.1 Terra is generally available in Copilot. GitHub listed the clients where the model shows in the picker. The list is Visual Studio Code, Visual Studio, the Copilot CLI and the Copilot cloud agent. It also runs on github.com, GitHub Mobile on iOS and Android, JetBrains IDEs, Xcode and Eclipse. GitHub said the rollout is gradual. GitHub also said the model is billed at provider list pricing under usage-based billing. Copilot Enterprise and Copilot Business plan administrators can turn access off through the model policy in Copilot settings.

Safety

OpenAI said the model meets the Critical threshold in cybersecurity under its Preparedness Framework. The company said it kept the protections developed for GPT-6 Astra. OpenAI said prompt-injection tests showed strong resistance to malicious instructions. The system card holds the detail.

What a team should do with this

The number worth watching is cost per useful task on your own evals. The headline price is not that number. Vendor benchmarks run with the model's prompts and the vendor's tool scaffolding. Your workload has different prompts, different tool calls, and a different cache hit rate. The honest comparison runs the model on your last 200 real traces. Measure the dollar cost, the latency and the pass rate on your own graders. A cheaper model that fails your eval suite is still too expensive. A model at half the price can still open new agent workloads. Only if it holds the same pass rate on your evals.

Source

• Primary source: OpenAI, "Introducing GPT-6.1 Terra".
• GitHub availability: GitHub Blog, "GPT-6.1 Terra in GitHub Copilot".