Local AI benchmark write-ups with methods and caveats.https://skrlr.com/https://skrlr.com/benchmarks/2026-10-27-distilled-video-drafts/https://skrlr.com/benchmarks/2026-10-27-distilled-video-drafts/Draft and final differ in steps, resolution, and frame count — not an apples-to-apples speedup multiplier.Tue, 27 Oct 2026 00:00:00 GMThttps://skrlr.com/benchmarks/2026-10-27-hunyuanvideo-fast/https://skrlr.com/benchmarks/2026-10-27-hunyuanvideo-fast/Fast variant only; 5 s output length; no batch; numbers from app logs, not a controlled harness.Tue, 27 Oct 2026 00:00:00 GMThttps://skrlr.com/benchmarks/2026-10-27-local-llm-decode/https://skrlr.com/benchmarks/2026-10-27-local-llm-decode/Decode only; prefill rate not published here; speed depends on speculative decoding acceptance.Tue, 27 Oct 2026 00:00:00 GMThttps://skrlr.com/benchmarks/2026-10-27-overnight-coding-agent/https://skrlr.com/benchmarks/2026-10-27-overnight-coding-agent/Different GPU class than video benchmarks; tests passed but product tasks did not.Tue, 27 Oct 2026 00:00:00 GMThttps://skrlr.com/benchmarks/2026-10-27-seedvr2-upscale/https://skrlr.com/benchmarks/2026-10-27-seedvr2-upscale/One representative job from logs; frame count and output resolution fixed for this run.Tue, 27 Oct 2026 00:00:00 GMThttps://skrlr.com/benchmarks/2026-10-27-shared-app-24h/https://skrlr.com/benchmarks/2026-10-27-shared-app-24h/One day; three people; not a load test; median wait is from the same window.Tue, 27 Oct 2026 00:00:00 GMThttps://skrlr.com/benchmarks/2026-10-27-text-to-image/https://skrlr.com/benchmarks/2026-10-27-text-to-image/Resolution and step count vary by workflow; range comes from app logs, not a fixed harness.Tue, 27 Oct 2026 00:00:00 GMT