Dev Probes 34 LLM APIs Daily, Catches DeepSeek Silent Changes

A developer has run daily fixed probes against 34 hosted LLM APIs from 15 labs since August 20, comparing each model only to its own history with pure-code scoring. The effort caught DeepSeek silently changing its models, with reasoning token usage jumping several-fold overnight.

2026-09-30 ~ 2026-10-01 · 2 related posts

• Daily probes across 34 LLM APIs catch DeepSeek reasoner's silent 10x token spike — Electrical_Rip892 · 2026-09-30
• Probing 34 LLM APIs daily since August caught DeepSeek's silent 3x cost jump on reasoner — Electrical_Rip892 · 2026-10-01