Tool-call failure rates jump 14x across providers, charts pinpoint inference stack updates
remilouf · x · 2026-11-12
Using dottxt's Tool-call Index data, the author asks: when did providers quietly update their inference stacks? Charts show tool-call failure rates stepping up sharply at specific timestamps — Together's qwen3.5-9b went from 0.19% to 3.2%, SiliconFlow's deepseek-v4-pro from 1.1% to 3.6%, some fixed in days, one took six.
Key takeaway: the same model can break agent tool calls 14x more often on the worst provider vs the best, and the breakpoints line up exactly with provider-side stack changes — a bigger reliability factor for agents than model choice itself.
More from coding & agent
- How to cut Fable 5.2 token costs by 90%: Arc routing plus a lean CLAUDE.md — EXM7777 · 2026-11-12
- Hamel Husain's complete AI evals framework for PMs and engineers — aakashgupta · 2026-11-12
- AI coding in one year: from unusable to full automation with no code review — angelaqdai · 2026-11-12
- Interactions Tool Searches Web Animations by Motion for Claude Code to Rebuild — yuwen_lu_ · 2026-11-12
- Agent Substrate Calculator Estimates Session Resume Speed Under Load — pjausovec · 2026-11-12
- Senate Bill Would Force Big Platforms to Accept AI Agents, Raising Auth Privacy Questions — Prof_CarlaReyes · 2026-11-12