OpenAI

Artificial Intelligence · AI Agents · Developer Tools

Launch

OpenAI launches GPT-6.1 Terra, a faster, 50% cheaper computer-use model

It signals a shift in the AI race toward cost-per-task efficiency for routine office work, rather than just chasing raw benchmark scores.

• OpenAI launched GPT-6.1 Terra on October 26, 2026, a mid-tier model for computer-use tasks, available immediately in ChatGPT Work, Codex, and via API at $1/$5 per million input/output tokens.
• Terra sits between OpenAI's low-cost Luna and premium Sol/Astra models, built to automate everyday office work like reading invoices, filling forms, and interpreting technical drawings.
• On OpenAI's Agents' Last Exam benchmark, Terra scored 15 percentage points higher than predecessor GPT-5.6 Terra at about a third of the cost, and matched flagship GPT-6 Astra at one-tenth the price per token.
• Terra's factual-error rate on OpenAI's internal benchmark fell from 53.6% to 16.7% at low reasoning effort, and its Terminal-Bench 4.0 coding score more than doubled to 50.3%.
• Independent testing by Artificial Analysis found Terra pareto-optimal for cost-to-intelligence at low and medium effort, with its hallucination rate on a private knowledge benchmark dropping from 88% to 38%.
• Terra launched the same day on Amazon Bedrock and in GitHub Copilot, showing how quickly third-party platforms now adopt new OpenAI models for enterprise and developer workflows.
• The release shows AI labs increasingly competing on cost-per-task efficiency for routine computer-use agents, not just raw intelligence scores, as cheap automation becomes the real battleground for enterprise adoption.