Local Q8 Showdown: DeepSeek-V4-Flash-Vision Finishes Tasks 2x Faster Than Qwen3.8
pabloodiablo · reddit · 2026-09-07
The author compares DeepSeek-V4-Flash-Vision (DSV4FV) vs Qwen3.8-Flash-Next (Q38FN), both at Q8KXL, as local daily drivers on 2x Strix Halo 128GB via USB-C 4 with Llama (RPC).
Key findings:
- Raw token generation: DSV4FV is 40% slower than Q38FN.
- Task completion: DSV4FV is 2x faster — a task taking Q38FN 25 minutes on medium took DSV4FV just 12 minutes, which the author attributes to fewer hallucinations and less getting stuck.
- Q38FN's 'xhigh' mode is unusable: a 25-minute task on medium failed to complete in 3 hours.
- DSV4FV on max finished the same task in 37-44 minutes.
- Behavior: Qwen3.8 over-interprets underspecified instructions, adding too much and bogging down in its own creativity.
Verdict: DSV4FV is the better choice for professional coding, though the author remains a big Qwen fan overall.
More from Infra
- SurfSense: an open-source local NotebookLM with MCP agents and a prompt-caching lesson — FurtiveMirth · 2026-10-23
- Meta researcher: horrifying infra, not talent, is why OSS can't catch closed frontier — A_K_Nain · 2026-10-23
- Jane Street and NVIDIA speed up Mamba-2 with custom cuTile kernels — blelbach · 2026-10-23
- LatentLoop: asymmetric compute allocation cuts looped-MoE training cost by 19% — pmttyji · 2026-10-23
- Same AI request emits 8.5x more CO2 in Texas than in Sweden — jlundberg · 2026-10-23
- SK Hynix Pulls Plug on Solidigm US IPO as Chip Stocks Slide, Shares Down ~30% This Quarter — rohanpaul_ai · 2026-10-23