Meta's Muse Out-Downloaded ChatGPT on a "Good Enough" Model

Meta's Muse agent beat ChatGPT in downloads, and it runs on a model nobody calls frontier. Brett checked the Play Store live on the show: Amazon Quick had about 50,000 downloads, Muse had 500,000. So does the best model still matter? The best one did get cheaper this week. Claude Opus 5.5 landed on Amazon Bedrock on launch day at $4 in and $20 out per million tokens, alongside GPT-6 Sol and Luna, while Kimi K3 took about 63 days to show up. Travers explains why Xiaomi's open-weights MiMo V2.6 Pro is now his default for paid API calls at $0.43 per million input tokens. On the AWS side: a guide to "tokenomics" and AI cost tracking on AWS, Lambda MicroVMs as isolated sandboxes for AI agents, new T8i burstable instances, and why a $1,000 credit makes us nervous about CloudWatch Omni pricing. Brett also sets up nightly AI code reviews with Kiro schedules and fights Amazon Quick sign-in. Plus OpenAI agents that used a University of Toronto link shortener as a message board, and Claude finding a CRISPR-like enzyme system. Would you trade the best model for one that's good enough at a tenth of the price? Tell us in the comments, and subscribe for a new episode every Wednesday.