Observed arrival · 2026-09-05
VeloBenchmark — per-token LLM benchmarking console
An open-source, single-binary console for measuring OpenAI-compatible LLM endpoints through streaming token timing, repeatable suites, concurrent loads, reports, and telemetry.
Field notes
The console derives measurements from streamed token timing and reconciles them with provider-reported usage counts when available. Its test builder covers prompts, exact context fills, fixed-shape requests, and vision steps, while concurrent workers share a step barrier so their snapshots can feed a combined throughput timeline. An OTLP/HTTP-JSON receiver can also convert serving-engine telemetry into the same session-report format. The page identifies the project as AGPL-3.0 and describes a single-binary distribution with zero runtime dependencies.
Observed signals
Read the marks
Editorial observations of this landing page, not a rating.
One card from the complete issue