DeepSeek V4 Flash Local Benchmark: MXFP4 Quantization Balances Speed and Top Scores
WonderRico · reddit · 2026-08-05
A Reddit user updated their local LLM benchmark with results for DeepSeek V4 Flash 0731.
Running the MXFP4 quantized version from Bartowski, the model achieved impressive speeds of 1K tokens/s for prefill and 90 tokens/s for generation using Dspark. The data shows this version scores the highest yet while maintaining extreme efficiency, though it notably lacks vision capabilities.
Related event: DeepSeek V4 Flash Benchmarks Leak, Outperforming Pro(3 posts)→
More from Infra
- 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
- Google open-sources AX, a Kubernetes-style orchestrator for billions of AI agent tasks — thisdudelikesAI · 2026-10-16