The Daily Model

There is a particular kind of work that nobody describes on a résumé. Reading an incoming invoice. Typing its details into accounting software. Filling out the form that unlocks the next form. The work is not hard. It is relentless. And for a long time, the cost of asking a language model to do it at volume made the math unworkable.

OpenAI announced GPT-6.1 Terra on October 26, 2026, and the headline number is not a benchmark score. It is a price cut. Standard API pricing is 50% lower than GPT-5.6 Terra and GPT-6.1 Sol—$1 per million input tokens, $5 per million output tokens (OpenAI). OpenAI frames this explicitly as making high-volume automation practical, citing the invoice-to-accounting-software pipeline as the example.

The model pairs that pricing with computer-use capability. On Agents' Last Exam, a benchmark of professional tasks that includes turning architectural drawings into 3D models and logging animated-film shots for restoration teams, OpenAI reports Terra scores 15 percentage points above GPT-5.6 Terra at medium reasoning effort, at roughly one-third the cost (OpenAI). OpenAI also claims Terra completes tasks over three times faster than both GPT-6 Astra and GPT-6.1 Sol. These are vendor-reported figures; no independent evaluation appears in the available material.

What interests me is the factuality claim. On OpenAI's internal benchmark of error-inducing prompts, the error rate at low reasoning effort dropped from 53.6% with GPT-5.6 Terra to 16.7% with Terra (OpenAI). That is a large reduction, and it matters precisely because the tasks in question—forms, invoices, data entry—punish a single wrong digit. A model that is convincing when wrong is a liability in that context. Still, this is an internal benchmark, and the prompts are OpenAI's own.

On Terminal-Bench 4.0, OpenAI reports Terra solves 50.3% of command-line coding tasks at maximum reasoning effort, more than double GPT-5.6 Terra's 21.5% (OpenAI). For the most complex coding work, OpenAI itself points users toward GPT-6.1 Sol or GPT-6 Astra. Terra is not positioned as the frontier. It is positioned as the workhorse.

The model is available now to Plus, Pro, Enterprise, and Edu users in ChatGPT Work and Codex, and via the API as `gpt-6.1-terra` (OpenAI).

The invoice problem was never about intelligence. It was about whether you could afford to run intelligence ten thousand times a day. That arithmetic has shifted. Whether the model holds up outside the benchmark—whether it actually reads the smudged invoice, the oddly formatted form, the PDF with the rotated table—remains the question developers will answer in production, not in an announcement.