OpenAI's GPT-6.1 Terra beats GPT-5.6 Terra by 15 points at one-third cost

GPT-6.1 Terra sits in the line of Terra models that OpenAI positions around practical work rather than pure benchmark chasing. The company's own framing emphasizes reading invoices into accounting software, matching product photos to catalogs, and operating apps to complete tasks, with speed and cost as explicit goals. The reported Agents’ Last Exam result (15 points above GPT-5.6 Terra at roughly one-third the cost) is the main data point OpenAI offers for that claim, and the API's 1.05 million-token context window suggests the model is meant to hold longer working documents and screens in view. Pricing is straightforward at $1 per 1 million input tokens and $5 per 1 million output tokens, with a surcharge once input passes 272,000 tokens, so heavier-context uses carry a different cost profile. Availability spans ChatGPT Work, Codex, and the API, with admin controls governing organizational access in the first two.