Today
-
Card: link the code on GitHub
-
Card: state model separation by head-to-head task wins (81 of 120 pairs at 11 or more of 12), with a grid
-
Card: 0 of 120 pairs is a limit of the corrected test at this size, not a finding about the models
Yesterday
-
Card: with no significant pair, say so (the sentence printed +nan)
-
Card: link the run viewer on oroagents.com
3 days ago
-
smb-bench-v1-3 append: summary
-
smb-bench-v1-3 append: traces/runs/gemini-4-hydrogen
4 days ago
-
smb-bench-v1-2 append: summary
-
smb-bench-v1-2 append: traces/runs/muse-spark-1-3
-
smb-bench-v1-2 append: traces/runs/muse-ember
-
smb-bench-v1-2 append: traces/runs/gpt-6-luna-high
-
smb-bench-v1-2 append: traces/runs/gemini-4-carbon
-
smb-bench-v1-2 append: traces/runs/gemini-4-argon
-
smb-bench-v1-2 append: traces/runs/claude-haiku-5-5
-
smb-bench-v1-2 append: traces/runs/claude-fable-5-2
6 days ago
-
Card: link the Harbor listing (oro-agents/smb-bench@1.0.0) and the lab access terms
-
Tag the oroagents.com link with UTM parameters for analytics
8 days ago
-
SMB Bench v1: final public-audit fixes (cards, Croissant, models/prices tables, GLM served id in traces)
-
recompute.py: compare escaped dollar values so --check passes on the card
10 days ago
-
Card: escape dollar signs so they are not rendered as LaTeX
-
SMB Bench v1 release