GPT-6.1 Terra: what to test before you replace GPT-5.6 Terra

GPT-6.1 Terra is OpenAI's new fast, efficient model for visual reasoning, computer use and writing, released on 26 October 2026. It costs less per token than GPT-5.6 Terra and keeps the same context window. Most calls switch with a model name change, but tool calls sent through Chat Completions must move to the Responses API first.

What OpenAI released

GPT-6.1 Terra was announced by OpenAI on 26 October 2026 in Introducing GPT-6.1 Terra, summarised as "bringing fast and capable computer use to everyone", and in the API changelog. The model ID is gpt-6.1-terra, and it is available on v1/responses and v1/chat/completions.

In the GPT-6 guide, Terra sits between GPT-6.1 Sol and GPT-6 Luna. OpenAI suggests it for matching product photos to catalogue entries, comparing technical drawings, acting on application screens, and drafting or revising text. The same day, GitHub made it generally available in Copilot for everyday agentic coding and terminal work.

What changed from GPT-5.6 Terra

Comparing the GPT-6.1 Terra model page with the GPT-5.6 Terra model page:

• Price: on OpenAI's pricing page, GPT-6.1 Terra's input rate is half the rate listed for GPT-5.6 Terra, and its output rate is less than half.
• Knowledge cutoff: 1 June 2026, up from 16 February 2026.
• Context and output: unchanged at 1,050,000 tokens of context and 128,000 maximum output tokens. Input is text and images; output is text.
• Reasoning effort: the same six levels, from none to max, with medium as the default.
• Positioning: GPT-5.6 Terra was described as the model that balances intelligence and cost. GPT-6.1 Terra is described for visual reasoning, precise computer use and natural writing.
• Tool calling: the GPT-6 guide says tool calling with GPT-6.1 Terra requires the Responses API. Chat Completions still works for plain requests.

It also picks up what the GPT-6 family adds: async tool calling, mid-turn steering over a WebSocket connection, and configuration_update items that change reasoning effort mid-conversation without breaking the prompt cache.

Pricing details that change your bill

The model page adds rules that matter more than the headline rate for agent workloads:

• Cache writes are charged. Cached input is billed at a twentieth of the uncached rate, but writing to the cache costs more than plain input, and 12-hour retention costs double the 30-minute rate.
• Long prompts cost more. Requests over 272K input tokens are billed at a higher rate for the whole request, input and output.
• Batch and Flex are half the Standard price; Fast mode is double; regional processing adds a premium where available.
• Data residency: the guide says Fast mode supports EU data residency for GPT-6.1 Terra.

Rate limits start at 5,000 requests per minute on the Build tier. The free tier is not supported.

Where teams can use it

On the API, it supports function calling, Structured Outputs and streaming. In the Responses API its tools include web search, file search, code interpreter, hosted shell, apply patch, skills, computer use, MCP and tool search. Fine-tuning is not supported. The model page lists one snapshot, gpt-6.1-terra, with no dated version to pin.

In GitHub Copilot, it is available to Pro, Pro+, Max, Business and Enterprise users in clients including VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, the cloud agent and github.com, with a gradual rollout. GitHub reports that in early testing it solved noticeably more terminal and C# coding tasks than GPT-5.6 Terra. It is billed at provider list pricing under usage-based billing.

What it means for teams building agents

CodeDTX's view: the price cut is the reason to test it, and computer use is the reason to test it carefully.

Cost. Output carries the higher per-token rate, so for an agent on GPT-5.6 Terra the output cut is where most of the saving sits. Measure cost per task rather than per token, because a model that takes more steps can erase a per-token saving. Our guide to controlling AI agent running costs covers how to set that baseline.

Computer use. OpenAI is pitching Terra at acting on screens. If you plan to let it click through an internal application, give it a sandbox and a test account first, and require approval before any action that writes, pays or sends. The same rules apply as for Claude's browser and computer use toolsets.

Copilot governance. GitHub says new models are enabled by default unless an administrator has turned off the global default or disabled this model. If your organisation reviews models before engineers use them, check the Copilot model policy today.

Choosing a tier. If you were weighing Sol against Astra, our post on GPT-6 Astra or GPT-6.1 Sol covers the top of the family. Terra now gives a cheaper step below Sol for screen and image work.

Migration steps from GPT-5.6 Terra

1. Run your evals at the same effort. The guide says to compare both models on representative images and workflows using the same reasoning effort. Start at medium and raise it only if the quality gain justifies the time and tokens. Use the set behind your AI agent evals.
2. Move tool calls to Responses. Any function calling still sent through Chat Completions needs to move before you switch.
3. Strip unsupported parameters. When reasoning effort is above none, remove temperature, top_p and top_logprobs, and logprobs on Chat Completions.
4. Recheck prompt caching. Look at cache hits and cache write charges in your usage data after the switch, not just the per-token rate.
5. Keep a fallback. Keep GPT-5.6 Terra configured behind a flag so you can switch back without a deploy, and plan for the day a model is deprecated.

When not to switch yet

Wait if your agent relies on tool calls through Chat Completions and you cannot schedule the Responses migration, if you need fine-tuning, which the model page lists as not supported, or if your regulator or customer contracts need a dated snapshot recorded against each release. Also wait if your evals show more steps per task on your own workflows, since that can cancel out the lower rate.

Frequently asked questions

What is GPT-6.1 Terra?

GPT-6.1 Terra is an OpenAI model released on 26 October 2026, with the API name gpt-6.1-terra. OpenAI describes it as fast and efficient, for visual reasoning, precise computer use and clear writing. It accepts text and image input, returns text, has a 1,050,000-token context window and supports reasoning effort from none to max. It sits between GPT-6.1 Sol and GPT-6 Luna.

Is GPT-6.1 Terra cheaper than GPT-5.6 Terra?

Per token, yes. OpenAI's model pages list lower input and output rates for GPT-6.1 Terra, with input at half the old rate and output at less than half. Cache writes, long prompts over 272K tokens and Fast mode change the real cost, so measure cost per completed task on your own workload before assuming the full saving.

Do I need to change code to move from GPT-5.6 Terra?

Usually only the model name, plus two checks. Tool calling with GPT-6.1 Terra requires the Responses API, so function calls sent through Chat Completions must move. When reasoning effort is above none, remove temperature, top_p and logprobs parameters. Then rerun your evaluation set at the same reasoning effort and compare quality, steps and cost against GPT-5.6 Terra.

Can I use GPT-6.1 Terra in GitHub Copilot?

Yes. GitHub made it generally available on 26 October 2026 for Copilot Pro, Pro+, Max, Business and Enterprise users, in VS Code, Visual Studio, JetBrains IDEs, Xcode, Eclipse, Copilot CLI, the cloud agent and github.com, with a gradual rollout. Business and Enterprise administrators control access through the model policy, and new models are enabled by default unless an administrator has turned that off.