News · Use it
AI SDK · openai-compatible 3.0.28
Clamps outputTokens.text at zero when a provider reports more reasoning tokens than completion tokens — seen on Baseten when a reasoning model hits the length stop mid-thought.4
Relevant to: anyone metering usage behind an OpenAI-compatible proxy serving reasoning models
What the verdict meansAvailable, works, worth your time today.
Sources
Originally part of the 10 August 2026 edition.