davanstrien/benchmark-diversity-atlas

Benchmark Diversity Atlas

Interactive embedding visualization comparing task diversity across three coding/agent benchmarks:

• —SWE-bench Verified (500 tasks) - Python open-source bug fixing
• —SWE-bench Pro (731 tasks) - Enterprise multi-language software engineering
• —Terminal-bench 2.0 (89 tasks) - Diverse terminal agent tasks

Built with Embedding Atlas. MCP endpoint available at /mcp for tool integration.