Model Watch

When a model ships, we re-test everything.

Within 48 hours of a notable release, the full task battery re-runs and each report answers one question: what changed for the work you actually do?

Specimen · a watch emailformat preview

Subject: Gemini 3.1 — 3 of 20 practitioner
tasks change verdict

Marketing copy: Gemini overtakes GPT
  (examples inside).
Long-doc summarization: unchanged, Claude holds.
Data extraction: improved 18% — if you gave
  up on Gemini, retry.

→ Your watched prompt ("LinkedIn post"):
  re-test on the new lineup in one click.

Keep this working

Get the next watch report.

Release-triggered only. No weekly filler — if nothing ships, nothing arrives.