AI News
OpenAI GPT-6.1 Terra: Half-Price Computer-Use Model at $1 Input and $5 Output Per Million Tokens
OpenAI released GPT-6.1 Terra on October 26, 2026, and the announcement leads with price. The mid-tier model costs $1 per million input tokens and $5 per million output tokens through the API — half the rates of GPT-6.1 Sol and, by OpenAI's own description, 50% cheaper than the GPT-5.6 Terra it replaces. Cached input costs $0.05 per million tokens.
Terra sits in the middle of OpenAI's lineup: above GPT-6 Luna, the cheapest model, and below GPT-6.1 Sol and GPT-6 Astra, the most capable. OpenAI describes it as a model that "pairs sharper visual reasoning and precise computer use with clearer, more natural writing," aimed at everyday office work: drafting proposals, reading technical drawings, filling out forms, entering invoices into accounting software and editing video.
Where you can use it
Terra rolled out on October 26 to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through the API as gpt-6.1-terra. Amazon Web Services made it available on Bedrock the same day. Two limits matter: Terra is not in regular Chat — you pick it in ChatGPT Work or Codex — and no free tier is listed, so API use is billed per token and product access needs a paid plan. OpenAI also notes that availability varies by plan, region, workspace settings and rollout progress.
The technical specs are generous. The context window is 1.05 million tokens with up to 128,000 output tokens, a knowledge cutoff of June 1, 2026, and text and image input. Tool calling requires the Responses API. One pricing detail catches long-prompt users: once a prompt passes 272,000 input tokens, the higher rate applies to the whole request — $2 input and $7.50 output per million. Batch and Flex processing cut standard prices by 50%, while Fast mode doubles them.
The benchmarks — OpenAI's own numbers
Every performance figure below comes from OpenAI itself; independent tests are not out yet, so treat them as the vendor's case, not a verified result.
On Terminal-Bench 4.0, a coding test, Terra solved 50.3% of tasks at maximum reasoning effort, against 21.5% published for GPT-5.6 Terra. On OpenAI's internal set of prompts designed to trigger errors, the error rate at low reasoning effort fell from 53.6% with GPT-5.6 Terra to 16.7% with Terra, and to 4.3% at maximum effort — nearly matching GPT-6 Astra and GPT-6.1 Sol. On Agents' Last Exam, a test of professional tasks, OpenAI says Terra scored 15 percentage points above GPT-5.6 Terra at roughly a third of the cost, and matched Astra at a tenth of the price per token. The company also says Terra completes tasks more than three times faster than Astra and Sol, a claim resting on its own timing runs.
What to keep in perspective
OpenAI itself points the hardest coding and research work to Sol or Astra, so Terra is a volume model, not a flagship: the one you run for the thousands of routine computer tasks, not the one hard problem. The model also meets the Critical cybersecurity capability threshold under OpenAI's Preparedness Framework, its highest tier — a classification worth noting for anyone letting a computer-use model click around real systems.
The honest summary: Terra halves the price of OpenAI's computer-use tier and posts large gains over its predecessor on OpenAI's own tests. Whether those gains hold up in independent testing, and whether the computer-use behaviour is reliable enough for unattended office work, is what the next few weeks will show.