DeepSeek V4 Flash on dual local GPUs hits ~1400 tokens/sec fixing iOS bugs
HankYeomans · x · 2026-09-02
A developer ran DeepSeek V4 Flash (MXFP4, 1M context) locally via LM Studio as a coding agent and fixed two iOS bugs in one unattended run.
- Claude's harness caps 'unknown models' at 200K context, so he wired it up another way
- No CI, linting, or Xcode tests were set up; a few well-placed skills let it plan and solve the bugs, with Amp reviewing the work
- LM Studio logs show 1400 tokens/sec, consuming nearly all VRAM on two GPUs while still leaving 1.5GB for the Wayland display
- He calls the quality comparable to frontier coding agents in pre-testing
Related event: Dev Runs DeepSeek V4 Flash Locally on Dual GPUs to Fix iOS Bugs(2 posts)→
More from coding & agent
- LLM one-shots a Minecraft 3x3 piston door by building its own Linux test infra — Vjeux · 2026-10-18
- Dev builds Mumbai-inspired 3D manga background generator from OpenStreetMap data — CurieuxExplorer · 2026-10-18
- Solo founder builds a money-printing data service on Claude's Pi Durable harness — MicahBerkley · 2026-10-18
- Dev builds Arch-based OS with custom desktop and browser using Claude Code — Snoo-10493 · 2026-10-18
- Five SaaS agent workflows fail the same way — and it's never model intelligence — usehive · 2026-10-18
- Claude skill generates hand-drawn interactive diagrams that teach kids how things work — CatAstro_Piyush · 2026-10-18