AgentX tutorial

Exploring agentic workloads: detailed telemetry

A single AgentX datapoint represents thousands of requests across growing conversations, subagents, warmup periods, cache states, and dynamically changing in-flight load. This is how to open one up.

Every card below opens the telemetry behind one AgentX configuration. Pick a model and a serving stack, and the detail page shows that run charted request by request.

per-point telemetry charts
11
views per point
3
replay stages
2
curve per model, SKU, and engine
1

AgentX runs with stored telemetry

141 serving configurations across 12 models, covering 5,172 benchmark points with per-request telemetry. Each card opens that configuration's headline point; the detail page's navigator moves between its concurrencies.

DeepSeek V4.1 Flash 552B

20 configurations · 1,161 points

Qwen 3.5 397B-A17B

19 configurations · 823 points

DeepSeekv4 Pro 0813 1.6T

17 configurations · 780 points

MiniMax M3 428B

14 configurations · 553 points

GLM 5.3 744B

15 configurations · 461 points

Kimi K3 2.8T

9 configurations · 422 points

MiniMax M3.1 Flash 428B

15 configurations · 381 points

Muse Spark Open 816B

10 configurations · 319 points

Qwen 3.8 Flash Next 176B-A6B

8 configurations · 133 points

Kimi K3.1 Pro 2.8T

10 configurations · 107 points

GLM 5.3 744B

3 configurations · 30 points

GLM 5/5.1

1 configuration · 2 points