RUNNER / PUBLIC
Bench board
Runner records deliberately reviewed and published by their owners. No combined speed/quality ranking.
Loading records…
Selected for comparison: 0/2
Legacy external lmx imports
BENCH / 01
Local LLM benchmarks
Check the conditions before comparing the numbers.
Loading records…
Different models, contexts and cache conditions do not share a speed ranking.Compare conditions →
Measurement reference · G16
Not a community recordRTX 5090 Laptop · 24 GB
Ultra 9 285H / RAM 64 GB · SC117 IQ3_XXS
90.3tok/s
Strata CUDA 0.1.50
reasoning-v1 · TTFT 2.15 s
reasoning-v1 · TTFT 2.15 s
An actual external local measurement. Not platform verified, user-published, or quality-scored.
Samples and limitations
70.6 / 94.6 / 90.3 tok/s
One warmup excluded, three measured requests, concurrency one. Per-request nonce prevented prefix reuse; cached tokens: zero. Each output ended at the 1,024-token length limit. This is not an answer-completeness, long-context, or energy-efficiency test.
Read the measurement guide →