What happened
OpenAI announced GPT-6.1 Terra on October 26, 2026, in a post on its Developer Community. The company describes it as a model that understands what is on screen, takes action across apps, and writes naturally. The announcement positions Terra as a more efficient option for computer use and a broad workday workload, rather than the choice for every demanding task.
The model is rolling out to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex. Developers can access it through the OpenAI API with the model name gpt-6.1-terra. OpenAI says its standard API prices are 50% lower than those of GPT-5.6 Terra. The announcement links to a model card for pricing, rate limits, and additional information, but the supplied material does not include the actual per-token prices or limits.
OpenAI’s performance claims center on computer use. It says GPT-6.1 Terra substantially outperforms GPT-5.6 Terra on Agents’ Last Exam, which it describes as a test of how well agents use software at work. It further says Terra matches GPT-6 Astra on that test at one-tenth the price per token and finishes tasks more than three times faster on average than both GPT-6 Astra and GPT-6.1 Sol. These are company-reported comparisons; the announcement excerpt does not provide detailed methodology or underlying results.
The company’s usage guidance draws a boundary around the model’s intended role: it recommends GPT-6.1 Sol or GPT-6 Astra for in-depth research or complex coding. For ChatGPT Work and Codex users, it says the usage allowance goes further with Terra than with GPT-5.6 Terra, and much further than with GPT-6.1 Sol or GPT-6 Astra. The post does not quantify those allowances or explain how they are calculated.
What the archive adds
This is a follow-up to our October 26 Field Note on GPT-6.1 Terra, which covers the same event. This announcement restates the launch and pricing/performance claims rather than establishing a separate release or new fact in the supplied evidence.
Trend signal
This supports agent economics moving to completed work: OpenAI frames Terra’s value through task completion speed and performance per token cost, not only model-call price. The evidence is a vendor’s benchmark claim, however, and does not establish cost per verified completed task in production.
Why now
OpenAI says the model is rolling out now and makes it available across ChatGPT Work, Codex, and the API. That simultaneous access matters because teams can evaluate it both as a product option and as an API model, though the source does not give a rollout schedule beyond “now.”
Connections to other events
An earlier channel post on August 17 described OpenAI lowering prices for GPT-5.6 Luna and Terra, including Terra’s then-current price. This announcement extends the price-and-efficiency thread to a new Terra version, but does not provide enough per-token figures here to compare absolute prices directly.
Why it matters
A lower-cost model that can operate software interfaces may make it more economical to run computer-use tasks repeatedly, especially where latency and usage limits constrain a workflow. Teams already using Codex or ChatGPT Work have a direct route to compare Terra with their current model, while API developers can test it in an existing integration under a distinct model name.
The useful decision is not whether Terra “matches” a larger model in the abstract, but whether it completes a team’s actual tasks reliably enough to offset failures, retries, and review. The announcement’s speed and price claims make that evaluation timely; they do not establish a general performance ranking. OpenAI itself points developers doing complex coding and research toward Sol or Astra, signaling that the cheaper tier may involve task-specific trade-offs. Earlier coverage: GPT-6.1 Terra Model: OpenAI API.
What happens next
- Teams evaluating computer-use agents will compare Terra with Sol or Astra on task completion cost, including retries and review. (3 months, 65%). Watch for: published evaluations that report end-to-end cost and success rates on comparable workflows.
Limits and uncertainty
The supplied announcement does not include per-token prices, API rate limits, benchmark scores, sample sizes, or the detailed protocol behind Agents’ Last Exam. “One-tenth the price per token” does not mean one-tenth the total cost of completing a task, and “over 3× faster on average” does not specify the task mix or comparison setup. The material also does not establish performance on teams’ own interfaces, failure rates, or whether lower ChatGPT usage consumption translates to a predictable allowance increase.
Skill to build now
Choose a repeatable computer-use workflow your team already runs, define what counts as successful completion, and record elapsed time, model/API spend, retries, and human corrections. Run the same task set with the current model and GPT-6.1 Terra before switching production traffic. This tests the announcement’s specific speed-and-cost proposition without treating a vendor benchmark as a guarantee for your environment.
Reader takeaway
GPT-6.1 Terra is available to try in ChatGPT Work, Codex, and the API, with a stated 50% standard-price reduction versus GPT-5.6 Terra. Treat OpenAI’s speed and benchmark comparisons as hypotheses: measure verified completion cost on your own computer-use tasks before deciding whether Terra can replace a larger model.