UCSB AI Benchmarks

RISE Leaderboard

Humanities data-extraction benchmarks (RISE, University of Basel) run against UCSB inference services. Every cell is normalized to a 0-100 scale, so columns are directly comparable.

Total Runs
58
Benchmarks
3
Models
28
Providers
3
Last Run
Sep 15, 2026
Provider
Vision
History
Showing 16 of 28 models โ€” sorted by Avg (high โ†’ low, missing runs last)
#ModelRunsAvg
0-100
Book Advert XML
Text XML ยท Fuzzy
Bibliographic Data
Printed List ยท Fuzzy
Fraktur Adverts
Fraktur OCR ยท CER
Last Run
๐Ÿฅ‡
AICommonsclaude-v5-opus๐Ÿ‘๏ธ vision
3
93.2
Sep 14, 2026
๐Ÿฅˆ
AICommonsclaude-v5-sonnet๐Ÿ‘๏ธ vision
3
88.3
Sep 14, 2026
๐Ÿฅ‰
AICommonsclaude-v4.6-sonnet๐Ÿ‘๏ธ vision
3
87.4
Sep 14, 2026
4
AICommonsclaude-v4.6-opus๐Ÿ‘๏ธ vision
3
87.0
Sep 14, 2026
5
AICommonsclaude-v4.5-haiku๐Ÿ‘๏ธ vision
3
78.7
Sep 14, 2026
6
CITgemma-4-31b๐Ÿ‘๏ธ vision
3
74.1
Sep 14, 2026
71
96.1
โ€”โ€”Sep 14, 2026
81
94.6
โ€”โ€”Sep 15, 2026
91
94.2
โ€”โ€”Sep 14, 2026
101
94.1
โ€”โ€”Sep 14, 2026
111
93.1
โ€”โ€”Sep 14, 2026
121
91.2
โ€”โ€”Sep 15, 2026
131
89.9
โ€”โ€”Sep 15, 2026
141
87.7
โ€”โ€”Sep 15, 2026
151
79.1
โ€”โ€”Sep 15, 2026
161
78.5
โ€”โ€”Sep 15, 2026
โš ๏ธ Flagged Models Excluded from Rankings (6)
These models were evaluated on RISE but scored a 0.0 average across all attempted benchmarks due to gateway token or request constraints (e.g., output token or context window limits on structured extraction requests) and are excluded from the ranked leaderboard and statistics above:
AICommonsamazon-nova-liteAICommonsamazon-nova-microAICommonsamazon-nova-proAICommonsllama-4-maverick-17b-instructAICommonsllama-4-scout-17b-instructAICommonsqwen3-32b
All cells share the 0-100 normalized scale; bars are absolute. โ€” means the benchmark was not run for that model.