OpenAI strengthens its agents with GPT-6.1 Terra

OpenAI launches GPT-6.1 Terra, a model designed for agent tasks, vision, and writing. The company promises better results than GPT-5.6 Terra at a significantly lower cost.

OpenAI launched GPT-6.1 Terra on October 26, 2026, a model presented as faster, more efficient, and more comfortable with agent tasks than GPT-5.6 Terra. The company targets proposals, technical drawings, forms, and video editing, with immediate availability in ChatGPT Work and Codex for Plus, Pro, Enterprise, and Edu subscribers. The message is clear: do better, faster, and if possible without blowing the budget.

Un modèle taillé pour les tâches du quotidien professionnel

Terra combines detailed visual reasoning, precise computer use, and more natural writing. OpenAI describes it as a clear upgrade over GPT-5.6 Terra, with better performance on concrete uses where models still sometimes falter: filling out a form, interpreting a diagram, editing a video, or drafting a cleaner proposal.

On Agents' Last Exam, OpenAI says Terra in medium reasoning mode scores 15 points higher than GPT-5.6 Terra, at about one-third the cost, while matching GPT-6 Astra at one-tenth the price per token. At maximum effort level, the company claims the model is over three times more token-efficient than Claude Opus 5.5 and charges about a quarter of the price per token, resulting in a cost per task about nine times lower.

Des gains de coût, mais pas pour tous les cas d’usage

OpenAI also emphasizes speed. Terra would complete tasks more than three times faster than GPT-6 Astra and GPT-6.1 Sol on average, which matters as much as raw score for agents expected to execute chained actions. On MMMU-Pro, the model outperforms GPT-5.6 Terra at all reasoning levels and approaches GPT-6 Astra and GPT-6.1 Sol at lower cost.

The company adds that on OSWorld 2.1, Terra at medium effort surpasses the best GPT-5.6 Terra score for a fraction of the cost per task. Similar trend on coding: on Terminal-Bench 4.0, the model solves 50.3% of tasks at maximum effort, versus 21.5% for GPT-5.6 Terra. OpenAI notes, however, that GPT-6.1 Sol and GPT-6 Astra remain preferable for the most complex coding work. In other words, Terra aims for the right balance, not the absolute throne.

Sécurité, disponibilité et prix : ce que change vraiment le lancement

OpenAI states that GPT-6.1 Terra reaches the Critical level of its Preparedness framework for cybersecurity and retains GPT-6 Astra's protections. The associated system card also mentions better compliance with guidelines, increased transparency on limits and blocks, and enhanced resistance to prompt injection in tests.

The model is accessible via API under the identifier gpt-6.1-terra. Regarding standard pricing, OpenAI announces 1 dollar per million input tokens, 0.05 dollar per million cached input tokens, and 5 dollars per million output tokens, which is 50% less than GPT-5.6 Terra and GPT-6.1 Sol. For companies already tracking costs closely, this difference is far from trivial.

By the numbers

  • 15 points — gain on Agents' Last Exam compared to GPT-5.6 Terra, medium mode.
  • 50.3% — tasks solved on Terminal-Bench 4.0, maximum effort.
  • 1 dollar — price per million input tokens, standard API.
  • 0.05 dollar — price per million cached input tokens.
  • 5 dollars — price per million output tokens, standard API.

To read