DeepSeek V4-Flash Slashes Compute to 66K GPU-hours Per 1T Tokens
teortaxesTex · x · 2026-08-01
A thread details the evolution of compute consumption across DeepSeek's models: dropping from 300K H800 GPU-hours per 1T tokens for V1, to 173K for V2, 180K for V3, and an estimated 66K for the latest V4-Flash.
For inference, V4-Flash and V2 have the same input/output prices, but cache hits are 5x cheaper and speed has roughly tripled or quadrupled. Their current best model is now smaller and cheaper than the version from 26 months ago.
Related event: DeepSeek Drastically Reduces Training Compute Costs Across Models(2 posts)→
More from Infra
- Grok claims 816B MoE model runs locally on dual M5 Ultra Mac Studios via MLX — grok · 2026-10-16
- Same 1,000 Cloudflare Durable Objects: $4,050/month cut to $22 with one trick — threepointone · 2026-10-16
- NVIDIA ships FP4-quantized Qwen3.8-Flash for local use, trending on Hugging Face — nvidia · 2026-10-16
- Baidu's Virtual Head Attention tames the KV cache, FLOPs, quality triangle in long-context LLMs — baidu · 2026-10-16
- DuoDiscover agents auto-build fast eval proxies, scoring 2.68x TPU kernel speedup — berkeley_ai · 2026-10-16
- Google Raises 2026 Capital Spending to Up to $205 Billion — emmanuelvivier · 2026-10-16