OpenAI has introduced GPT-6.1 Terra, a fast model that combines visual understanding, operating desktop applications and more natural writing. In the API it costs $1 per million input tokens and $5 per million output tokens, about 50% less than GPT-5.6 Terra. The company reports a sharp drop in factual errors and more than double the score on command-line coding tasks. The model is rolling out in ChatGPT for Plus, Pro, Enterprise and Edu users and in Codex.

What it means: The interesting part is not only the benchmark scores but the price. When a model can read documents, open forms and fill in fields in ordinary applications at half the cost, many small automations start to make economic sense. OpenAI itself gives the example of reading incoming invoices and entering them into accounting software, work that in Greece is tied directly to myDATA and mandatory e-invoicing. Before putting such a tool into production, though, controls are needed. Automated entries should be checked against rules, for example matching VAT numbers, totals and tax rates, and anything sent to the tax authority should have human sign-off. OpenAI also places the model at the top cybersecurity risk level in its own framework, so anyone giving it access to systems should limit permissions and log every action it takes. For Greek small and medium businesses, the sensible move is a small pilot on one repetitive task, measuring cost per task and error rate, before any wider rollout.

Read the original on OpenAI

Related service: Custom software

Related stories