OpenAI just cut the mid-tier computer-use invoice in half and kept the Critical cyber label on the box.
On 26 October 2026 the company introduced GPT-6.1 Terra as a fast visual and computer-use model aimed at high-volume desk work: invoices into ledgers, drawings into models, forms, video logs. The price is the Monday fact. The benches are the brochure that travels with it.
GPT-6.1 Terra lands at $1 / $5 with a Critical cyber tag
GPT-6.1 Terra is available to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and to developers via the API as gpt-6.1-terra. Standard API pricing is $1 per million input tokens, $0.05 per million cached input tokens, and $5 per million output tokens. OpenAI states those standard prices are 50% lower than GPT-5.6 Terra and GPT-6.1 Sol.
On Agents' Last Exam, a professional computer-use suite that includes turning architectural drawings into a 3D model and shot-by-shot film logging, OpenAI reports GPT-6.1 Terra at medium reasoning effort scoring 15 percentage points above GPT-5.6 Terra at about one-third the cost, and on par with GPT-6 Astra at about one-tenth Astra's price per token.
At max effort, Terra is described as over three times as token-efficient as Claude Opus 5.5 and one-fourth its price per token, for roughly 9x lower cost per task, and more than three times faster than GPT-6 Astra and GPT-6.1 Sol on the tasks cited.
MMMU-Pro visual reasoning improves over GPT-5.6 Terra at every reasoning setting and nearly matches GPT-6 Astra and GPT-6.1 Sol at lower cost, per the post. OSWorld 2.1 desktop-agent scores at medium effort surpass GPT-5.6 Terra's best at a fraction of the cost per task.
Terminal-Bench 4.0 at maximum effort reaches 50.3% versus 21.5% for GPT-5.6 Terra; OpenAI still points complex coding to Sol and Astra. Terminal-Bench-Science 0.1 at max effort is said to beat GPT-6.1 Sol on cost and come within five points of Opus 5.5 at less than one-fifth estimated cost.
On an internal error-inducing prompt set, low-effort error rate falls from 53.6% (GPT-5.6 Terra) to 16.7%; max-effort error rate falls 18 points to 4.3%, nearly matching Astra and Sol. Safety: GPT-6.1 Terra meets the Critical threshold in cybersecurity under OpenAI's Preparedness Framework and keeps GPT-6 Astra protections. Alignment notes claim better scope honesty and prompt-injection resistance versus GPT-5.6 Terra, with detail deferred to the system-card addendum published the same day.
API fleets, Work seats, and anyone still on 5.6 Terra pricing
OpenAI shipped the weights-as-a-service SKU and the price card. Teams on ChatGPT Work and Codex with Terra-class usage inherit the new allowance economics. API customers automating invoice capture, form fill, or visual QA sit closest to the $1 / $5 math. Safety and security reviewers inherit the Critical cyber designation for routing and monitoring policy. Anthropic Opus 5.5 appears only as a vendor cost-efficiency foil on OpenAI's page.
Half-price Terra with Astra-parity claims on Agents' Last Exam
A dated mid-tier computer-use model at half the prior Terra list price, with OpenAI claiming Agents' Last Exam parity to GPT-6 Astra at roughly one-tenth token price, doubled Terminal-Bench 4.0 versus GPT-5.6 Terra, and a retained Critical cyber Preparedness threshold. Distinct from prior Sol/Astra monitoring and offline-training containment stories: this is a SKU and price-path ship.
Harnesses, regions, and independent benches stay vendor-side
The launch post and same-day system-card addendum are dated 26 October 2026. Standard list is $1 / $5 per million tokens, with cached input at $0.05; the page describes API-hosted access and does not offer local weights. Availability is stated for Plus, Pro, Enterprise, and Edu Work/Codex paths; country carve-outs beyond product norms are not itemised here, and use sits under hosted API terms.
Agents' Last Exam, OSWorld 2.1, Terminal-Bench 4.0, MMMU-Pro, and the internal factuality set are OpenAI-reported. Third-party reproduction under disclosed harnesses remains unsettled on this announcement.
Re-price high-volume computer-use jobs; keep Critical routing
Platform owners running invoice, form, or visual-agent loops on GPT-5.6 Terra should re-quote unit economics at $1 / $5 and A/B gpt-6.1-terra against current traces before locking Q4 automation budgets. Security reviewers should map the Critical cyber threshold into the same monitoring tier used for Astra-class tools. Treat Agents' Last Exam parity language as a vendor score until an independent suite prints the same comparison.