SemiAnalysis Open Sources AgentX 1.0: $3M Agentic Coding Benchmark

woosuk_k · x · 2026-08-25

SemiAnalysis released AgentX 1.0, an open-source multi-turn agentic coding benchmark built from $3M of real traces on 1000+ chips. vLLM reported competitive performance, with DeepSeek V4 Pro achieving 130k tok/s/chip. The release covers vLLM's optimizations for prefix reuse, long context parallelism, and scaling with PD disaggregation.

Related event: SemiAnalysis Open-Sources $3M AgentX Benchmark for AI Coding Agents(2 posts)→

Original post →

More from coding & agent

coding & agent channel →