UCSB AI Benchmarks

Operational QoS

These metrics characterize the end-user experience of each provider's infrastructure — time to first token, generation throughput, and inter-token latency — not model capability.

Total Runs
1698
Benchmarks
14
Models
52
Providers
4
Last Run
Oct 31, 2026
Provider
Metric
Filter
History
Showing 37 of 52 models — sorted by Avg Gen TPS (overall) (higher is better)
#ModelRunsOverall
Gen TPS
advanced math
Gen TPS
chemistry
Gen TPS
computational law
Gen TPS
computer science
Gen TPS
digital humanities
Gen TPS
mathematics
Gen TPS
neurobiology
Gen TPS
philosophy
Gen TPS
scientific coding
Gen TPS
signal processing
Gen TPS
Last Run
🥇33Sep 30, 2026
🥈48Oct 31, 2026
🥉45Oct 30, 2026
445Oct 30, 2026
545Oct 30, 2026
645Oct 30, 2026
730Sep 4, 2026
845Oct 30, 2026
945Oct 30, 2026
1045Oct 30, 2026
1145Oct 30, 2026
1245Oct 30, 2026
1333Sep 30, 2026
1430Oct 1, 2026
1530Oct 1, 2026
1630Sep 5, 2026
1745Oct 30, 2026
1830Oct 1, 2026
1930Sep 5, 2026
2030Oct 1, 2026
2133Sep 30, 2026
2233Sep 30, 2026
2345Oct 30, 2026
2430Sep 5, 2026
2545Oct 30, 2026
2645Oct 30, 2026
2730Oct 2, 2026
2845Oct 30, 2026
2930Sep 5, 2026
3033Sep 30, 2026
3133Sep 30, 2026
3233Sep 30, 2026
3345Oct 30, 2026
3433Sep 30, 2026
3533Sep 30, 2026
3645Oct 30, 2026
3733Sep 30, 2026
Bars show relative performance within the visible selection. Click a score cell to expand the runs behind it.