OpenAI has filled the gap in its GPT-6 lineup. GPT-6.1 Terra, announced on October 26, 2026, is a mid-sized model built for computer work, and it costs half as much as the model it replaces. Input tokens run $1 per million and output tokens $5 per million, down from $2 and $10 for GPT-5.6 Terra, with cached input at just $0.05 per million tokens.
Terra was the missing piece. When OpenAI launched the GPT-6 family in September, it shipped the flagship Astra, the larger Sol, and the small Luna, but no Terra. The name had already surfaced in Microsoft evaluation configurations before the announcement, which hinted that a mid-tier model was on the way. Now it is live in ChatGPT Work, Codex, and the API under the identifier gpt-6.1-terra, for Plus, Pro, Enterprise, and Edu users.
Built for workday computer tasks, not chat demos
OpenAI positions Terra as a fast, efficient model that combines detailed visual reasoning with precise computer use and clearer, more natural writing. The pitch is practical: drafting proposals, reading technical drawings, filling out forms, and editing video. That is the unglamorous middle of office work, and it is where most AI spending actually happens.
The benchmark claims back up the positioning. On OpenAI’s Agents Last Exam, Terra scored 15 points higher than its predecessor, matched the flagship Astra model at one tenth the token cost, and finished tasks more than three times faster. On a private knowledge test, its guessing rate fell from 88 percent to 38 percent while accuracy rose. On OpenAI’s internal benchmark of error-inducing prompts, the error rate at low reasoning effort fell from 53.6 percent to 16.7 percent, and at maximum effort it fell 18 percentage points to 4.3 percent.
Terra undercuts Claude Sonnet on price per token
The pricing table is where Terra gets interesting. At $1 input and $5 output per million tokens, it costs less than half of Anthropic’s Claude Sonnet 5.5, which charges $2 and $12, and it sits squarely between OpenAI’s own Luna and Sol in the lineup.
| Model | Input per million tokens | Output per million tokens | Best for |
|---|---|---|---|
| GPT-6 Luna | $0.10 | $0.50 | Cheap, high-volume tasks |
| GPT-6.1 Terra | $1 | $5 | Computer use and office work |
| GPT-6 Sol | $2 | $10 | Heavier reasoning |
| Claude Sonnet 5.5 | $2 | $12 | Coding and long-form work |
| GPT-6 Astra | Flagship pricing | Flagship pricing | Maximum capability |
For Indian developers, the headline rate works out to roughly Rs 88 per million input tokens at current exchange rates, before taxes and platform fees. Cached input at $0.05 per million tokens is the quieter bargain, since repeated prompts over the same documents get dramatically cheaper.
Independent testing puts Terra on the cost frontier
Third-party testing supports at least part of the claim. Artificial Analysis calls GPT-6.1 Terra a new benchmark for intelligence per dollar. At maximum reasoning effort, the model gained 6 points on the Intelligence Index over GPT-5.6 Terra while costing less than a third per task, $0.40 versus $1.40. At lower effort levels the gains were larger, 7 to 13 points, placing the model on the Pareto frontier, the point where cost and performance can no longer both improve at once.
That framing matters because it matches the pattern of the last month. OpenAI has been winning the AI coding war with discounts rather than benchmarks, cutting prices to pull developers off rival models. Terra extends the same strategy to computer use, the next battleground after coding.
What this means if you build with AI
If you already use GPT-5.6 Terra, the switch is straightforward and the savings are immediate: same tier, half the price, better scores. If you use Sonnet for agent-style work, Terra is now the cheaper default to test against, though you should benchmark it on your own tasks before migrating, since OpenAI’s internal tests flatter its own models.
The bigger signal is the direction of the market. Our October token price list showed mid-tier models converging on a few dollars per million tokens, and Terra pushes that floor lower still. Meanwhile DeepSeek’s latest coding model is squeezing the same tier from the open-weight side. The flagship race gets the headlines, but the mid-tier price war is where most developers’ bills are decided.
One caveat: OpenAI notes Terra does not currently support UAE inference residency, so teams with data residency requirements should check availability before building on it. For everyone else, the model is available now, and the price cut is real money from day one.
