Summary
IndyDevDan uses OpenRouter’s public data to break down the state of AI models heading into the final quarter of 2026. The video is organized around five trends, starting with the surge in token usage. OpenRouter’s weekly volume has grown from about 4.5 trillion tokens a year ago to 146 trillion last week, which the host cites as evidence against the idea that AI demand is a bubble.
That growth prompts a discussion of “tokenomics,” the idea that more tokens only matter if they produce more useful work. The host also examines Anthropic’s reported plans for a $2 trillion IPO despite large losses, and what would need to hold true for its projected revenue growth to materialize. He notes that OpenRouter does not count tokens from direct provider usage, which explains why few Anthropic models appear on its leaderboard.
A major segment covers Typesafe’s Jev, a zero-shot classifier model that has climbed to 12th place on OpenRouter in its first week while maintaining strong uptime. The host argues it is harder to replace with clones than many assume and shows early benchmarks of Jev inside his Pi coding agent alongside Sonnet 5.5, saving roughly 20 percent on cost for a sample prompt. Engineers weighing speed, cost and performance trade-offs will find useful context here.
📺 Source: IndyDevDan · Published October 05, 2026
🏷️ Format: News Analysis
