Benchmark category
Published agents evaluations and source-native model rankings. Each benchmark keeps its original scale and methodology.
Data as of 2026-08-07
Select a benchmark to inspect model-level results, evidence fields and scoring direction.
| BrowseComp | BrowseComp | text | Score | 58 | featured | B | Yes |
| Terminal-Bench 2.0 | Terminal-Bench 2.0 | text | Score | 49 | featured | C | No |
| SWE-Bench Pro | SWE-Bench Pro | text | Score | 44 | featured | B | No |
| Toolathlon | Toolathlon | text | Score | 31 | featured | B | Yes |
| MCP Atlas | MCP Atlas | text | Score | 30 | featured | B | Yes |
| Finance Agent v2 | Finance Agent v2 | text | Score | 26 | featured | B | No |
| Terminal-Bench | Terminal-Bench | text | Score | 25 | featured | B | No |
| t2-bench | t2-bench | text | Score | 23 | featured | B | Yes |
| OSWorld-Verified | OSWorld-Verified | multimodal | Score | 22 | featured | B | No |
| OSWorld | OSWorld | multimodal | Score | 20 | featured | B | No |
| BFCL-v3 | BFCL-v3 | text | Score | 19 | featured | B | Yes |
| DeepSWE 1.1 | DeepSWE 1.1 | text | Score | 19 | featured | B | No |
| Terminal-Bench 2.1 | Terminal-Bench 2.1 | text | Score | 17 | featured | B | No |
| FrontierCode 1.1 | FrontierCode 1.1 | text | Score | 15 | featured | B | Yes |
| FrontierSWE | FrontierSWE | text | Score | 15 | featured | B | No |
| NL2Repo | NL2Repo | text | Score | 14 | featured | B | No |
| BFCL-V4 | BFCL-V4 | text | Score | 13 | featured | B | No |
| Claw-Eval | Claw-Eval | text | Score | 13 | featured | B | No |
| Legal Agent Benchmark | Legal Agent Benchmark | text | Score | 13 | featured | B | No |
| CyberGym | CyberGym | text | Score | 11 | featured | B | No |
| VITA-Bench | VITA-Bench | text | Score | 10 | featured | B | Yes |
| DeepSWE | DeepSWE | text | Score | 10 | featured | B | No |
High-coverage benchmarks with at least two published model results.
How this category is assembled on llmboard.ai.
This page groups benchmarks whose primary or display category matches agents. It does not average incompatible metrics into a new category score.
Open an individual benchmark to inspect score direction, evidence level, participant count and source-native results.
Category membership is derived from the current benchmark registry response.