OPEN LINK ↗
// 1h agoBENCHMARK RESULT
llm-speed publishes measured local LLM benchmarks
llm-speed is an open-source leaderboard for measured local LLM inference speed across models, hardware, and runtimes. Its signed runs report decode throughput, prefill speed, time to first token, and reproducible configuration details.
// ANALYSIS
The valuable contribution is turning scattered “my GPU gets X tok/s” claims into auditable, comparable measurements.
- –Fixed workloads and pinned model digests make cross-machine comparisons more meaningful
- –Reporting decode, prefill, and TTFT avoids reducing real inference performance to one misleading number
- –Ed25519-signed submissions improve provenance, though they do not independently verify hardware or output quality
- –The open CLI and CC BY dataset make the project useful for hardware buying guides, local deployment planning, and reproducible research
// TAGS
llm-speedllmbenchmarkinferenceopen-sourceclilocal-first
DISCOVERED
1h ago
2026-10-19
PUBLISHED
1h ago
2026-10-19
RELEVANCE
7/ 10
AUTHOR
Syntax