GPT-6 Astra vs Sol: breaking down the launch price sheet line by line
The same job costs $7.50 on GPT-6 Astra and $1.50 on Sol, while Astra Ultrafast runs to $45, and the post walks through the published rates to explain the 5x and 30x gaps.
OpenAI just undercut its own flagship by 80%. The same launch price sheet has a tier that costs 30 times as much for an identical job. Here is one job run through every line of it. What the flagship side charges, as announced: > $10 - a million input tokens on GPT-6 Astra > $50 - a million output tokens on Astra > $60 - that same million in on Astra Ultrafast > $300 - that same million out on Ultrafast > $500 a month - the Pro 500 tier where Ultrafast reportedly sits Now take a job of 500k tokens in and 50k out. Calculated from those prices, it's $7.50 on Astra. GPT-6.1 Sol does it for $1.50. Same job, 5x cheaper, and launch coverage says Sol matched the flagship on DeepSWE 1.1, a coding benchmark. The rest of the bill, line by line: > INPUT: $2 per million tokens on Sol, against $10 > OUTPUT: $10 per million, against $50 > CACHED INPUT: $0.10 per million, 95% off Sol's normal input price > CONTEXT: a claimed 1.05M-token window (a launch recap lists 1M), with higher rates above 272,000 input tokens. Filling it once is about $2.10 of input. On Astra that's $10.50, before any surcharge > ERRORS: factual mistakes in answers reportedly fell from 11.4% to 7.7% > ULTRAFAST: up to 8x faster token generation in Codex, a quoted 6x in the API, at six times Astra's standard price Ten such jobs a day save about $60 a day on Sol versus Astra, roughly $1,800 over 30 days. Run the same job on Ultrafast and it costs $45, thirty times the Sol bill. The $45 buys speed. At 8x, a 40-minute generation finishes in 5 minutes at best. A walkthrough of the launch with 314,546 views on X warns that the gain stops at token generation, so the minutes spent waiting on a browser or a tool call stay where they were. It also says extra credits on cheaper plans don't unlock Ultrafast. The old rule of model pricing said the best answers cost the most. On this sheet the most expensive line sells speed, and the $2 line reportedly matched the flagship on a coding benchmark. The whole bill gets recalculated in the video ↓
This post quotes another post: x.com/joyingeth/status/2106050928042238314