ResearchAudio

Evidence-led AI briefings and free decision tools for engineers, researchers, and product builders.

Latest

Pathway's 150M Model Never Writes Its Reasoning Down

Aug 19, 2026

29.5% on ARC-AGI-1 for 0.07 cents per task. The failure map is the best part.

Qwen's 27B Can See. Its 2.4T Flagship Cannot.

Aug 17, 2026

Alibaba's first Max-class open weights arrive with no vision stack, thinking locked on, and a license that names its own products.

The Guardrail on GLM‐5.3 Is a Calendar

Aug 15, 2026

Z.ai shipped its strongest cyber scores and held the weights for two weeks of hardening. The UK's security institute has already measured what hardening survives on open weights, and what a window of time is worth.

Grok 4.6 Is a 2 Dollar Model Until Token 200,001

Aug 13, 2026

Grok 4.6 Is a 2 Dollar Model Until Token 200,001

The Best Skill Never Read a Reasoning Trace

Aug 12, 2026

Microsoft handed fifty ordinary agent logs to Claude Code. It compiled them into about a hundred lines of markdown that recover, and twice beat, an entire reasoning mode.