Action-native model skips JSON parsing, outputs action probabilities for cheaper, faster inference
FrankFelixAI · x · 2026-09-16
A repost highlights an "action native" model: instead of sequentially generating JSON text that gets parsed into actions like conventional LLMs, it directly outputs probabilities over discrete actions.
The author notes the exact architecture is unknown, but this specialization appears to make the model very cheap and fast to run — especially on the output side, since no autoregressive text decoding is needed. The approach reflects a broader trend of bypassing language tokens to model actions directly for agents and robotics.
More from coding & agent
- Cloudflare open-sources security-audit-skill for coding agents with machine-readable findings — cloudflare · 2026-09-16
- Tencent's open-source WeKnora RAG knowledge platform tops 24.8k GitHub stars — Tencent · 2026-09-16
- Supabase keeps climbing: the open-source Postgres platform for AI apps hits 109k stars — supabase · 2026-09-16
- Cline: open-source autonomous coding agent ships as SDK, IDE extension, or CLI — cline · 2026-09-16
- Anthropic open-sources knowledge-work plugins built for Claude Cowork, hits 24k stars — anthropics · 2026-09-16
- Building a real-time Gemini Live-style AI voice and vision app in iOS with SwiftUI — amos_gyamfi · 2026-09-16