gemini-4POST
Gemini 4 Argon early access: 99% of figures match across 250+ evals
A team with early access to Gemini 4 Argon ran 250+ evals, scoring each cell against the cited filing figures, and found 99%+ matched, with frontier results on 22 of 25 hardest tasks.
Vinoo Ganesh
@VinooGaneshArchived 20 days agoOriginally posted 2026.10.14
Vinoo Ganesh @VinooGaneshOriginally posted 2026.10.14
We had early access to Gemini 4 Argon from @GoogleDeepMind and ran it through 250+ evals, each scored cell by cell against figures cited to the filing that reported them. 99%+ of figures matched. Frontier on 22 of our 25 hardest tasks. Shoutout to @gl3nnleblanc for leading the work!
This post quotes another post: x.com/gl3nnleblanc/status/2110043072801169611
x.com/vinooganesh/status/2110056736560398571Archived 2026.10.14 10:17