DeepSeek-V4.1-Flash on four DGX Sparks, single stream, live 🐋
The predecessor (V4 Flash + DSpark) peaked around 50 to 70 tok/s on code. V4.1 is 552B total, 16B active on the way out, 8B active on prefill, and about half the checkpoint is n-gram memory tables that we keep on
This campfire cost $0.10.
One prompt, 41 minutes, 1,398 lines. DeepSeek v4.1 Flash — Three.js, TSL, WebGPU, procedural fire and sound. The agent checked its own work: headless Chrome screenshots caught 3 rendering bugs, CDP confirmed the audio graph was actually running. 60
DeepSeek V4.1 Flash (High) vs Kimi K3 (Low)
Deepseek-v4.1-flash on DeepSeek-V4.1-Flash TP4
· 6 parallel coding agents • Total: 5,046 tokens in 36s • Aggregate: 140.9 tok/s sustained · 138.4 wall-avg · 242.8 peak burst • Per-stream decode: 25.9 high / 20.1 low / 23.5 avg tok/s • TTFT 0.63s avg · E2E 36s avg
You have reached the end of the archive
All of deepseek