AgentX
Also known as AgentX benchmark, AgentX scenario
In plain English
AgentX is the InferenceX workload for testing how inference systems serve complete long-context, multi-turn coding-agent sessions.
Technical definition
AgentX is InferenceX’s agentic inference benchmark scenario, built from workload shapes derived from opt-in coding-agent traces after original content is removed.
Engineering details
AgentX reconstructs session structure with deterministic synthetic tokens. Its replay keeps request lengths, turn timing, shared-prefix growth, tool pauses, and main-agent or subagent dependencies while excluding original prompts, generated code, and tool payloads. The serving stack receives the traffic pattern without receiving the source conversation.
Why it matters
Long contexts pressure KV-cache capacity, repeated prefixes reward effective cache reuse, and branch timing tests request scheduling. These effects are small or absent in short, independent requests. The resulting curve describes the complete serving system under agent traffic.
How to read it in InferenceX
The Agentic scenario appears by default for models with matching AgentX data. Compare its throughput, latency, and interactivity only with other AgentX runs at compatible settings. Use fixed-sequence scenarios when the target workload is a conventional request stream.
Source material
See the concept in real benchmarks
Ultra-High Interactivity on NVIDIA GPUs? TileRT on InferenceX
Can TileRT software on NVIDIA GPUs compete with Cerebras, Groq LPU, and SambaNova? Batch size 1, disaggregated engine, high-throughput prefill engine, high-interactivity decode engine
Vera Rubin NVL72 vs GB200 NVL72? Inference TCO & Architecture Analysis
Rubin LUT Based Tensor Core, Feynman, Rack Scale, Perf Per MegaWatt, Perf Per Dollar, Software Improvements, Public Rubin Software, PyTorch, vLLM, OpenAI Triton