The rollout comes after OpenAI abandoned plans for an upgraded GPT-6.1 Astra model because the version frequently ignored instructions and proved unreliable. Now available to all Plus, Pro, Business, Enterprise, and Education users via ChatGPT Work and Codex, the new model features a factual error rate within 1.9% of Astra. For high-volume developers, OpenAI reduced cached input costs to $0.10 per million tokens, a 95% discount over standard input pricing.
Tuesday, October 20, 2026
Story
OpenAI's GPT-6.1 Sol Hits 75.2% on Coding Test after Scrapped Astra Update
AI21 days ago
OpenAI has deployed its GPT-6.1 Sol model to GitHub Copilot, Cognition's Devin, and its internal API, pricing the mid-range tool at $2 per million input tokens and $10 per million output tokens. The model outperformed its predecessor on the DeepSWE v1.1 agentic coding benchmark with a 75.2% success rate, compared to 68.8% for GPT-6 Sol, while reducing the cost per task by 76%. It also scored 71.4% on the OSWorld 2.0 offline set, approaching the 73.5% score of the Astra flagship at approximately one-seventh of the cost.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.