Atlas

Benchmarks

← All benchmarks

MCP Atlas public (DeepSeek-V4 report, unspecified protocol)

Agents

MCP Atlas public-set pass rates printed in the DeepSeek-V4 preview report and model cards. The report does not say whether the original or the April 2026 revised judge was used, and its baseline values mix figures matching vendors' launch tables with others, so the cohort is kept as a report-specific diagnostic.

Top models (higher is better)

ModelScore
DeepSeek-V4-Pro74.2
Opus 4.673.8
GLM-5.171.8
Gemini 3.1 Pro Preview69.2
DeepSeek-V4-Flash69.0
GPT-5.467.2
Kimi K2.666.6
Loading Atlas data…