0
Applied AI·August 13, 2026·1 min read

11X cheaper than ChatGPT: Tiny 150M model just proved AI doesn't need to "think out loud" to be smart

Share

A 150M model hitting 29.5% on ARC-AGI-1 at 11x lower inference cost than a comparable ChatGPT reasoning tier is a concrete data point that chain-of-thought isn’t the only path to useful reasoning. For teams cost-gated on API usage, it’s time to re-open the model menu and benchmark small models against your real tasks, not just leaderboards.