OpenAI says GPT‑6.1 Terra cuts error rate to 16.7% and is widely available

Published on Dorsal: October 30, 2026 at 9:04 AM (UTC)

Editorial update: October 30, 2026 at 10:34 AM (UTC)

Source date: October 27, 2026 at 12:40 PM (UTC)

AI-generated conceptual illustration for “OpenAI says GPT‑6.1 Terra cuts error rate to 16.7% and is widely available”. — AI-generated illustration

OpenAI presented GPT‑6.1 Terra and says it keeps the safety protections of GPT‑6 Astra, improves visual understanding and desktop task handling, and is available via ChatGPT Work, Codex and the API under the name gpt-6.1-terra; OpenAI published internal benchmark results and a per‑token pricing schedule for the API.

Summary

AI-assisted summary of the linked source

According to OpenAI and reporting in 20minutos, GPT‑6.1 Terra reduces failure rates in internal tests from 53.6% to 16.7% on deliberately misleading prompts, matches or exceeds some predecessor performance on coding and scientific-agent benchmarks, and is offered to Plus, Pro, Enterprise and Edu users plus developers via an API priced at one dollar per one million input tokens and five dollars for output tokens.

Go to source