OpenAI Cuts Computer-Use API Prices in Half With GPT-6.1 Terra: $1 Per Million Input Tokens, Terminal-Bench 4.0 at 50.3%

OpenAI released GPT-6.1 Terra on October 26, filling in the mid-tier of its model lineup with a system built for the kind of work that fills an office day: reading screenshots and technical drawings, filling in forms, drafting documents, and operating desktop applications. The company describes it as "a fast, efficient model that combines detailed visual reasoning and precise computer use alongside clearer, more natural writing," and as a significant step up from GPT-5.6 Terra, which it replaces.

The headline change is price. Standard API access to GPT-6.1 Terra costs $1 per million input tokens and $5 per million output tokens, with cached input at $0.05 per million. That is 50 percent below what OpenAI charged for GPT-5.6 Terra, and also below GPT-6.1 Sol. Developers call the model gpt-6.1-terra. With the release, OpenAI's lineup now runs Luna, Terra, Sol and Astra from cheapest to most capable — Terra missed last month's GPT-6 refresh, and this update brings the middle tier back in line with the rest of the family.

Access is immediate for Plus, Pro, Business, Enterprise and Edu users inside ChatGPT Work and Codex — the desktop app, Work on the web, Codex CLI version 0.159.0 or later, and the IDE extension. It does not appear in the regular Chat tab. OpenAI says the same usage allowance covers more tasks with Terra than with Sol or Astra, and that some Plus and Pro users will see it become the default model in Work's presets.

The benchmark numbers are, as always with a launch post, OpenAI's own and not independently verified. On Agents' Last Exam, a suite of professional computer-use tasks such as turning architectural drawings into 3D models, Terra at medium reasoning effort scored 15 percentage points above GPT-5.6 Terra at about one-third the cost, and performs on par with flagship GPT-6 Astra at one-tenth of its per-token price. OpenAI says Terra completes tasks more than three times faster than both Astra and GPT-6.1 Sol.

Coding and reliability improved as well. On Terminal-Bench 4.0 at maximum reasoning effort, Terra solves 50.3 percent of tasks against 21.5 percent for GPT-5.6 Terra. 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 dropped 18 points to 4.3 percent. OpenAI adds that Terra's token efficiency at maximum effort is more than three times that of Anthropic's Claude Opus 5.5 at roughly a quarter of its per-token price, which the company computes to about nine times lower cost per completed task — again, its own math. For the hardest coding work, OpenAI still points users to Sol or Astra.

On safety, OpenAI says Terra meets the Critical threshold in cybersecurity under its Preparedness Framework, keeps the protections built for GPT-6 Astra, and shows stronger resistance to prompt injection along with a better habit of admitting its limits rather than bluffing. Details sit in the system card.

The release lands in a market where computer use is no longer a demo. Anthropic, Google and Microsoft are all pushing agents that click through real software, and the binding constraint has shifted from raw capability to cost per task and tasks per hour. By cutting the price of its mid-tier in half while claiming near-flagship task performance, OpenAI is making the argument that most everyday agent workloads do not need the most expensive model — and pricing rivals' mid-tiers into a uncomfortable spot. The claims are worth treating with caution until independent evaluations catch up, but the direction is clear: the per-task economics of computer-use agents just moved.