OpenAI GPT-6.1 Terra Halves the Price of Getting Work Done

GPT-6.1 Terra Price Cuts the Cost of Everyday AI Work

OpenAI released GPT-6.1 Terra on October 26, and the headline is the GPT-6.1 Terra price: half what the previous Terra model charged through the API. Input tokens now cost $1 per million, cached input costs $0.05 per million, and output costs $5 per million. That is a direct cut from $2 and $10 for input and output on GPT-5.6 Terra. For teams that run AI on thousands of invoices, forms, or support tickets a day, that difference adds up fast. It also puts Terra well under Anthropic's Claude Sonnet 5.5, which charges $2 for input and $12 for output, according to price comparisons published alongside the launch.

OpenAI is not pitching Terra as its smartest model. It is pitching it as the practical one. The company says Terra is built for high-volume work: reading invoices and entering data into accounting software, drafting proposals, interpreting technical drawings, filling out forms, and editing video. That list matters because it describes jobs, not benchmarks. A model that is slightly less brilliant but twice as cheap wins when the task repeats a thousand times a day.

What Terra Actually Does Better

According to OpenAI, GPT-6.1 Terra improves on GPT-5.6 Terra in communication and computer-based tasks. It pairs visual reasoning with more precise computer control, so it can look at a drawing or a form and act on it rather than just describe it. Early third-party testing reported by AI Daily Post backs up part of the claim: Artificial Analysis called Terra a new benchmark for intelligence per dollar, gaining 6 points on its Intelligence Index over GPT-5.6 Terra at maximum reasoning effort while costing less than a third per task, $0.40 versus $1.40.

Availability is narrower than a normal ChatGPT launch, and that is worth knowing before you go looking for it. Terra is available in ChatGPT Work and Codex, plus the API as gpt-6.1-terra, for Plus, Pro, Business, Enterprise, and Edu users. It is not in regular ChatGPT chat conversations, and free users will not see it. Some Plus and Pro users may also see a Medium effort setting as a new default in Power presets, plus a Standard versus Fast mode toggle where supported. If your model picker looks different from a colleague's, that gradual rollout is probably why.

Why a Cheaper Model Matters More Than a Smarter One

The AI industry spent two years racing on capability. The current race is on cost per task. Claude Haiku 5.5 costs a tenth of its predecessor, and DeepSeek just shipped open weights with a cheap search API. Terra fits the same pattern: vendors know that most business AI work is routine, and routine work lives or dies on unit economics. A finance team automating invoice processing does not need a model that can write poetry. It needs one that reads a messy scan correctly, costs pennies per document, and does not need a human to redo every third result.

That is also why OpenAI framed the release as a shift in the price-to-performance curve rather than a straight upgrade. If Terra holds up in real deployments, the practical effect is that smaller companies can afford automation that used to be priced for enterprises only. A 20-person firm can run the same invoice pipeline as a 2,000-person firm, just at lower volume. The moat shifts from who can afford AI to who designs the workflow well.

What to Watch Before You Switch

Price cuts deserve a test, not a migration on day one. Run Terra on a sample of your real documents, not a demo set, and measure accuracy on the messy cases: handwritten notes, low-resolution scans, forms with missing fields. Check where it fails, because a model that is cheap and wrong at scale is more expensive than a dearer model that gets it right. Also confirm your plan and workspace settings, since Business and Enterprise admins may control availability, and Terra is not available with UAE inference residency.

For developers, the API pricing is simple enough to budget: $1 per million input tokens, $0.05 cached, $5 per million output. Cache aggressively if your prompts repeat system instructions or reference documents, because cached input is now almost free. And keep an eye on Moonshot's Kimi K3.1 Pro open weights, due by November 4, if you want a self-hosted alternative to compare against. Terra's real test will not be launch-week benchmarks. It will be whether teams still use it in three months, when the novelty fades and the invoice pile does not.