Listed rates versus actual task cost
- The introductory Gemini rate quoted is $2 per million input tokens and $10 per million output tokens. Theo compares that with $5/$25 for Opus 5.5, but argues that the lower unit price hides heavier token use.
4:31 - In the Artificial Analysis Intelligence Index results he shows, Gemini scores above Astra and 6.1 Soul but below Opus 5.5 High. Its benchmark run also costs more than Opus High. He treats this as a general intelligence indicator, with limited relevance to coding.
3:374:06 - Theo reports cache hits of about 70–80%, compared with 99%+ for other models in his usage. Missing the quoted 95% cache-read discount means paying more full-price input cost.
4:53 - He does not establish the cause. Possibilities include Antigravity request construction, usage reporting, or T3 code's integration, described here in the transcript as MCP. Treat the cache figures as an observation of his setup.
5:24
Scheduled price change reported in the video
- Theo says the introductory rates will double on December 1 to $4 per million input tokens and $20 per million output tokens. The passage supplies no year; these are the video's announced rates, not a current pricing check.
9:4710:11 - His planning objection is that introductory pricing followed by increases makes a model harder to build a business around. He would prefer sustainable launch pricing followed by reductions.
10:1110:32