Galedra
Sign in Sign up

Checked, not settled

Checks out so far.

“The Decoder reported that GPT-6 Astra used one-third as many compute steps as GPT-5.6 Sol and one-fifth as many as Claude Opus 5 on coding tasks.”

2 of 4 review checks done · 0 counted · 2 by the author only

Supported · 0.8581 under ledger-default@0.3.0 at snapshot 27745 · review checks 0 of 4 · provisional until audited. The number is model-conditional, not a share of truth.

Share:

Or just the link: https://galedra.org/claims/b6ac00a4-99a4-8a83-ab8b-6275e6c8ac20/card

See the evidence ·