GPT-6.1 Sol: Nearly as Smart as Astra at a Fraction of the Cost
Trending Topics Jakob Steinschaden ● Covered by 15 sources
OpenAI killed its Astra update and shipped GPT-6.1 Sol instead. It’s almost as smart, but much cheaper, which is exactly the fight now.
Based on reporting by Trending Topics, Jakob Steinschaden — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
OpenAI needed something new to show at DevDay after halting GPT-6.1 Astra, so it put GPT-6.1 Sol on stage instead. The pitch is simple: near-Astra performance for coding, computer use and office work, but at a much lower bill.
Artificial Analysis says the numbers mostly hold up. On its Intelligence Index, Sol scores 52 points at maximum reasoning effort. That is one point behind Astra and four points ahead of GPT-6 Sol, which arrived only a week ago. Among 222 models, Sol lands in 10th place.
The cost gap is the real story. Running the index with Sol comes out at $0.72 per task, versus $3.26 with Astra. OpenAI’s own API pricing sets Sol at $2 per million input tokens and $10 per million output tokens, while Astra sits at $10 and $50. The cache is where Sol gets especially cheap: processed input costs $0.10 per million tokens, 95 percent less than standard input.
On some tasks, Sol even gets a clean win. In the Coding Agent Index at the second-highest reasoning level, it beats Astra by one point while costing less than 15 percent as much. It also improves on Terminal-Bench, Humanity’s Last Exam and GDPval-AA, while the AA-Omniscience hallucination rate drops from 60 percent to 54 percent. That is better, but still means the model makes things up more than half the time when it does not know.
The wider market reaction says almost as much as the model card. On Hacker News, the launch went straight to the top with more than 850 points, and the thread quickly turned into a price-war discussion. Anthropic’s Claude Opus 5.5 and Sonnet 5.5 still lead the ranking, but Sol is cheaper on a per-task basis than either of them. OpenAI did not beat Anthropic on raw quality here. It did something more practical: it made the argument about cost, and that is where the pressure is now.
My take — AI-written commentary, not fact-checked reporting
This is what an AI price war looks like when the glossy demos wear off. OpenAI is no longer selling the smartest thing in the room; it’s selling the model that won’t empty the meter quite as fast. That is probably the right move, and also a sign the industry has started behaving like an electricity business with better branding.
Read more about this at: Trending Topics