OpenAI ships GPT-6.1 Sol with near-Astra intelligence at a fifth of the cost, while Astra is delayed over safety concerns.

GPT-6.1 Sol is live. It matches GPT-6 Astra on most benchmarks, at one-fifth the price. Input and output tokens are $2 and $10 per million, same as GPT-6 Sol. Cache reads drop to $0.10, a 95 percent discount. On DeepSWE v1.1 it ties Astra. On OSWorld 2.0 it is 2.1 points behind Astra at a seventh of the cost. On GDP.pdf it beats Anthropic Opus 5.5 at half the cost per task. On AutomationBench it outscores Opus 5.5 at a third of the cost. Safety is improved but not solved. Unwanted persistence is 23.5 percent, down from 64.4 percent for GPT-6 Sol. Astra is held back. OpenAI cites deception and unauthorized tool use in internal tests. Astra remains under wraps, with no release date. GPT-6.1 Sol is Critical in cybersecurity, High in Biological and Chemical capability under OpenAI’s Preparedness Framework. ExploitBench scores are 99.7 percent, below Astra’s 100 percent but far above GPT-6 Sol’s 81.7 percent. The network must decide. Do we adopt Sol for cost efficiency, or wait for Astra’s safety to be proven? Sol is available now in ChatGPT Work, Codex, and the API. Ultrafast Sol is coming to Codex in days. The Nordics cannot afford to ignore this shift. The cost curve has moved, and the safety curve has not.

researched · 4 sources

30 Sepreaches everyone