Salesforce's 3-Year xLAM Bet on Function Calling Ships as Koa
huan__wang · x · 2026-09-18
A Salesforce AI Research lead recounts how a 3-year-old bet — that teaching models function calling would matter enormously for agents — shipped this week as Salesforce Koa, its first reasoning model. The xLAM Large Action Model family was fully open-sourced along the way: models from 1B to 141B on Hugging Face, the 60K function-calling dataset, the APIGen data pipeline, APIGen-MT-5k, AgentOhana, and three papers.
More from coding & agent
- User has Muse AI agent call and book a dentist appointment in under 10 minutes for free — ChrisUniverse · 2026-09-18
- AI agent Jev beats Pokémon's first gym after 8,000+ decisions for $1.21 — HankYeomans · 2026-09-18
- LAX launches to bridge natural math language and Lean, but looks a lot like existing tool span — lpachter · 2026-09-18
- GitHub's gh secure: one command enables all free security settings on public repos — 0xkarasy · 2026-09-18
- GitLab hosts Kimi K3, GLM 5.3 and MiniMax M3, up to 4x calls per credit — RealGeneKim · 2026-09-18
- Demos vs. Reality: Coding Agents Stuck in Loops Burning API Budgets — Particular-Swan2404 · 2026-09-18