Worst inference provider breaks 19x more tool calls: dottxt launches provider quality index

remilouf · x · 2026-11-20

dottxt's new Tool-call Index shows the inference provider matters as much as the model: on the same model, the worst provider breaks up to 19x more tool calls than the best (e.g. gpt-oss-120b ranges from 0.01% to 71.96% broken calls; GLM 5.3 from 0.01% to 17.81%). The index covers 64 models, 66 providers and 15 model labs, with weekly reports.

remilouf also shipped an agent skill that lets your agent pick the provider itself:

npx skills add https://providers.dottxt.ai --skill choose-provider

Related event: dottxt Tool-call Index: Worst Providers Show 19x More Broken Tool Calls(3 posts)→

Original post →

More from coding & agent

coding & agent channel →