As OpenAI et al. ship Jev-style APIs, one dev trains a 149M router model for $6.60
MaziyarPanahi · x · 2026-10-01
With OpenAI, Liquid and Nace shipping Jev-style APIs this week, the author trained his own tool-routing model with a single prompt in HuggingFace's HuggingChat: an ML Intern run finished 6K training decisions in 16 minutes on one NVIDIA A100, fixing its own bugs along the way. The full 60K-decision version cost about $6.60 and picks the right tool 62% of the time out of 20 tools. Prompt, recipe and model card are public.
Related event: Dev Trains 150M-Param Router Model with a Single Prompt(2 posts)→
More from coding & agent
- Longtime Dev: "AI Can't Build Maintainable Architectures" Is a 2023 Take — dreamwieber · 2026-10-01
- Economist Paul Novosad: AI-edited code becomes unmanageable, so I wall it off — paulnovosad · 2026-10-01
- Local models can now power computer use agents, but regulated industries still lack a playbook — Ambitious_Fold_2874 · 2026-10-01
- Xiaomi MiMo-V2.6 report: one mixed GRPO run across all domains, $2.6M for Pro — SergioPaniego · 2026-10-01
- Synentra: an open-source gateway that lets AI agents' actions be judged by intent and risk before execution — ziagham · 2026-10-01
- Let agents write code to call tools instead of exposing every tool directly — Future_AGI · 2026-10-01