← All benchmarks

MCP Atlas (LongCat-2.5 report, unspecified protocol)

Agents

MCP-Atlas pass rate that Meituan ran for its LongCat-2.5 launch table, at reasoning effort "max", temperature 0.7, top_p 0.95 and 256K context. The post does not establish the task subset, judge or its revision, or the turn and tool-call budget; its starred comparator values match Scale's 1,000-task leaderboard, which does not make LongCat's own run comparable. Retained as a source-specific diagnostic.

Benchmark source: Longcat

Top models (higher is better)