Cost per run US dollars Long read $4.38 $3.49 Debug 1 $0.63 $0.52 Debug 2 $0.31 $0.28 Taste $0.17 $0.15 AE script $0.87 $1.70 Rules $0.26 $0.36 Image check $0.27 $0.24 Python $0.23
Are these our graphs? Must be our graphs not anthropic
13 Matching Annotations
- Last 7 days
-
pages.definitelyreal.com pages.definitelyreal.com
-
-
B
Paragraph break.
-
him
remove this word
-
The job: write the status note and review post for a shot.
Italic, with a paragraph break afterwards.
-
It
Fable 5.2 did not claim
-
A taste answer
We've trained agents to predict the director's notes, and judged questions of taste on how well they matched the director's pick.
-
the
remove
-
a 210,000-token log to read and answer questions on, two bugs we had already fixed, an After Effects script, a status post that had to follow the house rules, a review board with known mistakes, a small Python function, and one question of taste.
sub-bullet points
-
You have probably read the benchmark tables already. This is not one. It is
We did 67 bounded tests, comparing tasks given to Fable 5.1 in the past week. We controlled for inputs, ensuring that each had access to the same xyz (say anything about the process here)
-
It also
Which one did?
-
finishing
we are finishing, Reality Games.
-
. Ninety minutes later it was running on our machines and a different model was testing
, and right away we had Opus 5.5 test
-
against
vs
-