Tencent Hy4 Flash hits 45.5 tok/s on one DGX Spark; GLM-5.3 Flash 137 tok/s on 2 Macs
AIFlow_ML · x · 2026-11-20
AIFlow aggregates community benchmarks from 19-20 Nov: Tencent's Hy4 Flash (220B, 12B active) ran at 45.5 tok/s single-user on one DGX Spark with a drafter (124.2 tok/s at 8 users), scoring 76% on a 75-task reasoning eval at roughly half the tokens of Ripple 65B. Tencent's 4-bit Spark build hit 92% vs 91% for DeepSeek V4.1 Flash on three Sparks and 89% for GLM-5.3 Flash. On Macs, GLM-5.3 Flash reached 137 tok/s single-user on 2× M5 Ultra over TB5 with drafts on (89.5 on one), and 260 tok/s total across 32 users with zero errors in a 30-min soak. Numbers are labeled MEASURED-BY-SOURCE, not re-run.
Related event: Local benchmarks reveal Hunyuan Hy4 Flash speed and long-prompt load times(2 posts)→
More from Infra
- Hands-on: AMD ROCm 10.2 delivers 2.2x token throughput on Qwen3.8-27B — AnushElangovan · 2026-11-21
- OpenVINO: the open-source toolkit for running AI on Intel CPUs, GPUs and NPUs — abhishek__AI · 2026-11-21
- AWS Lambda now supports GPUs via EC2 capacity providers — samgoodwin89 · 2026-11-21
- 27,262 thinking tokens, zero answer: Google auto-billed 23 empty Vertex AI crashes via HTTP 200 — Plenty-Jackfruit-218 · 2026-11-21
- Hy4 Flash reads 1M tokens in 44 min on Mac Studio M3 Ultra, 81 min on DGX Spark — AIFlow_ML · 2026-11-21
- local-ai-registry: one battle-tested local LLM recipe per GPU, with evidence attached — LiTianleli · 2026-11-21