October 26, 2026, OpenAI officially released GPT‑6.1 Terra. Previously absent from some GPT‑6 series updates, Terra's return had been speculated against, but it now returns as a "fast, efficient mid‑tier" and is described by the company as a significant leap over the previous GPT‑5.6 Terra.
The new model emphasizes fine‑grained visual reasoning, precise computer operation and more natural writing ability, focusing on drafting plans, interpreting technical drawings, filling forms, editing videos and other high‑frequency office scenarios. It is positioned as the "workhorse" model for everyday workflows rather than a flagship seeking extreme capabilities.
Positioning and core capabilities
OpenAI's model lineup is roughly divided into Luna (lightweight), Terra (mid‑tier), Sol and Astra (higher‑end). GPT‑6.1 Terra is defined as a middle choice balancing speed, cost, and practicality. The company says it achieves coordinated improvements in visual understanding, screen operation, and text generation, able to more naturally understand screen content, execute cross‑application actions, and produce clearer, more work‑appropriate text.
Typical applications include:
- Reading and interpreting technical drawings or product images;
- Performing multi‑step operations in desktop applications (form filling, data entry, simple editing, etc.);
- Drafting and revising office documents;
- Acting as an agent to continuously perform computer tasks in daily work.
OpenAI emphasizes that compared with GPT‑5.6 Terra, the new version shows clear improvements on multiple internal and public benchmarks, and standard API prices have been reduced by 50%.
Performance and cost data (official disclosures)
According to OpenAI's published data:
- On the Agents’ Last Exam (which assesses an agent's ability to use software to complete professional tasks), at medium reasoning intensity, GPT‑6.1 Terra is about 15 percentage points higher than GPT‑5.6 Terra, and single‑task cost is roughly one‑third of the latter; in some settings, its performance approaches GPT‑6 Astra, while token price is about one‑tenth of Astra.
- On Terminal‑Bench 4.0, at maximum reasoning strength, task completion rate increased from GPT‑5.6 Terra's 21.5% to 50.3%.
- Fact error rate at low reasoning intensity dropped from 53.6% to 16.7%, and at maximum intensity further dropped to 4.3%, approaching flagship levels.
- Task completion speed is on average over three times faster than GPT‑6 Astra and GPT‑6.1 Sol.
- API pricing is $1 per million input tokens and $5 per million output tokens (cached inputs lower), about 50% lower than the standard prices for GPT‑5.6 Terra and GPT‑6.1 Sol.
It should be noted that the above figures mainly come from OpenAI's own tests and internal benchmarks; independent third‑party verification is still needed. For complex coding and deep research scenarios, the company still recommends using higher‑end Sol or Astra.
Availability and product rollout
GPT‑6.1 Terra is immediately available to ChatGPT Plus, Pro, Enterprise and Edu users within ChatGPT Work and Codex, and developers can call it via the API as gpt‑6.1‑terra. The context window is about 1.05 million tokens and supports text and image inputs. It supports multiple reasoning‑strength settings (including none) and continues tool capabilities such as computer use.
Under the ChatGPT Work and Codex quota systems, the same quota can cover more tasks than the previous‑generation Terra and significantly more than Sol/Astra, further reinforcing its positioning as the "everyday workhorse."
Objective view: reconfirming the mid‑tier market
There are several notable points in this release:
First, OpenAI has not abandoned the mid‑tier product line. While flagship models keep raising capabilities and prices, Terra's return shows the company still values "good enough, fast enough, cheap enough" workflow scenarios. For many enterprises and individual users, the frequent needs are stable screen operations and document tasks rather than always calling the most expensive model.
Second, price and efficiency are clear selling points. With API costs and agent invocation volumes rising, a 50% price cut and faster task completion directly lower the barrier to large‑scale deployment. This responds to previous industry concerns that agent bills might far exceed chat costs.
Third, capability boundaries still require caution. Although official data show clear progress, complex reasoning, long‑term autonomous tasks, and high‑risk scenarios are still better suited to higher‑end models. Improvements in vision and computer operation also increase demands for safety and controllability—especially regarding cross‑application operations and prompt‑injection protection.
Conclusion
The launch of GPT‑6.1 Terra is a pragmatic fill in OpenAI's model stratification strategy. It does not attempt to overturn the flagship narrative but focuses on "high‑frequency, scalable, and cost‑controlled" office and agent scenarios, trading lower price for more practical visual understanding and computer‑operation capabilities. For users and companies needing large‑scale automation of daily work, this may be more practically valuable than simply pursuing the strongest model.
Of course, the ultimate effect still depends on stability, error rates and third‑party verification in real scenarios. As AI moves from "able to answer questions" to "able to continuously operate computers," whether mid‑tier models can truly become the workhorse of workflows will be a focus to watch in the coming months.
▶️ Like this piece? Creating is hard!
Follow our YouTube channel:
AI Rosetta
Your likes, subscriptions and shares are our greatest support! 🙏
More AI news at:https://www.airosetta.com