OpenAI's GPT-6.1 Terra: $1/$5, half of 5.6 Terra, now in Codex
Built for high-volume computer use and automation, it tops OSWorld, ships in ChatGPT Work and the API, and meets OpenAI's 'Critical' cyber threshold.
Copy markdown
Half the price, and Flex halves it again
Standard runs $1/M input and $5/M output — 50% under GPT-5.6 Terra — with cache reads at $0.05/M and a 1.1M-token context. The async Flex tier drops to $0.50/$2.50, so batch and offline jobs get cheap fast.
The coding score more than doubled
On Terminal-Bench 4.0 at max effort it solves 50.3% of tasks, up from 21.5% for GPT-5.6 Terra. That makes it a credible cheap swap-in for CI and code-review agents — test it before you keep paying flagship rates.
It clicks through real desktop apps
Terra beats GPT-5.6 Terra's best OSWorld 2.1 (desktop-control) score at a fraction of the cost per task. Think cheap agents that read invoices, fill forms, match product photos, and drive GUIs you can't reach by API.
~9x cheaper per task than Opus 5.5
OpenAI claims that at max effort Terra is 3x+ more token-efficient than Claude Opus 5.5 at a quarter the per-token price — roughly 9x lower cost per task, and about 1/10 the price of GPT-6 Astra. Numbers are OpenAI's own, so verify on your workload.
Get it today — but not on the free tier
Live now as gpt-6.1-terra via Responses, Chat Completions and Batch, and rolling out in Codex and ChatGPT Work for Plus, Pro, Enterprise and Edu. The API free tier doesn't include it.
A 'Critical' cyber-risk model
Terra ships at the 'Critical' cybersecurity threshold of OpenAI's Preparedness Framework, carrying GPT-6 Astra's safeguards plus stronger prompt-injection resistance. Expect firmer refusals on security-adjacent prompts.