Models
Canonical model identities, providers, specifications, current pricing and provenance.
- Updated
- 1 Sept 2026
- Records
- 848
Lumina Bench · Research infrastructure
Structured model, benchmark, operational and Arena datasets used by LuminaBench.
Start with the full workbook, bundle or machine-readable manifest.
Value presence and audit resolution are separate. Conflicting, stale and unresolved records remain visible rather than being hidden inside a single completeness percentage.
Cohort: current public Top 100. The complete per-field and per-cohort breakdown is included in the ledger coverage JSON and workbook.
Canonical identities and exact evidence rows remain separate, stable and source-qualified.
Canonical model identities, providers, specifications, current pricing and provenance.
Exact reasoning and endpoint configurations retained separately from base models.
Benchmark identities, categories, lifecycle, weighting and source metadata.
Every public-eligible benchmark observation, including configuration-specific evidence.
Long-form pricing, performance, offering, surface and media endpoint observations.
Source-native and Arena systems stay separated by route, track and pool.
Image generation, editing, I2I-Bench and Arena evidence kept by leaderboard and track.
Text-to-video and image-to-video evidence kept by leaderboard and track.
Text-to-speech, voice cloning and speech-to-speech evidence.
Use these files to trace records and find missing or stale fields before a future refresh.
Namespaced production, refresh, media and Arena provenance metadata.
Model, benchmark, media and stored source-health coverage in one operational report.
Existing schemas remain available: CSV/JSON Models & providers, CSV/JSON Benchmarks, CSV/JSON Cost and CSV/JSON Speed & latency.