ModelCap
Current AI model comparison
Exact identities · public evidence
Qwen
Qwen3 Omni 30B A3B Instruct
Position
#92
Index
47.5
Estimated
publisher-corpus-prior over 186 held-out anchors (10%); launch card against 4 resolved peers on 4 rows (90%); exceeds every named peer on 1 of 4 rows; no cross-lab optimism probe available; shrunk 0.3 up toward the measured corpus
Comparison side 1
StepFun
Step 3.7 Flash
Position
#91
Index
47.7
Estimated
3 reported rows against measured corpus ladders (14%); global-corpus-prior over 186 held-out anchors (20%); measured predecessor stepfun/step-3.5-flash less the succession penalty (66%); shrunk 1.1 up toward the measured corpus
Comparison side 2
Output / 1M
Unavailable
vs $1.15
Context
—
vs 262K
Weights
Restricted license
vs Open weights