| Terminal-Bench 2.1(Vendor-measured (MiniMax internal infra, Claude Code scaffolding)) | 66% |
| SWE-bench Pro(Vendor-measured (MiniMax internal infra, Claude Code scaffolding)) | 59% |
| NL2Repo-Bench(Vendor-measured (MiniMax internal infra, Claude Code scaffolding)) | 42.1% |
| BrowseComp(Vendor-measured (MiniMax internal infra, Claude Code scaffolding)) | 83.5% |
| MCP Atlas(Vendor-measured (MiniMax internal infra, Claude Code scaffolding)) | 74.2% |
| BankerToolBench(Vendor-measured (MiniMax internal infra, Claude Code scaffolding)) | 76.1% |