TIGER-AI-Lab
We track 4 benchmarks that TIGER-AI-Lab made or helped make.
1 in use 4 with public code 4 with a leaderboard here
Benchmarks
| Name | Kind | Released | Best score | Status |
|---|---|---|---|---|
| StructEval Models generate or convert structured outputs in 18 formats, such as JSON, TOML and HTML. StructEval-T covers text-only formats and StructEval-V renderable ones. | Data science and SQL | May 2025 | | In use |
| WebsiteBench Agents explore reference websites and rebuild their appearance, behavior, and persistent state. Hidden browser tests require functional correctness and screenshot similarity, then report the fraction passed. | Web and app front ends | 26.2% | Not in cards | |
| ClawBench V2 Browser agents do 130 new everyday tasks on live websites under six supported harnesses. Request interception and a more lenient LLM judge score each run. | Web browsing | May 2026 | 44.6% | Not in cards |
| ClawBench Browser agents do 153 everyday tasks on live websites, such as booking travel, ordering food and applying for jobs. Request interception stops real side effects, and an LLM judge scores the five-layer traces. | Web browsing | Apr 2026 | 61.4% | Not in cards |