FOSS stack with DeepSeek beats paid ChatGPT plan in 2-month agent reliability test
layer4down · reddit · 2026-10-22
A developer ran three voice-driven, hands-off agent workflows side by side for two months: ChatGPT+Astra, DeepSeek+Qwen+MLX, and Claude Code+GLM, all wired to a private GitHub repo in 24/7 loops.
- Reliability ranking: DeepSeek > Claude Code > ChatGPT. DeepSeek (v0.7.1-rc1) ran near-flawlessly for days on a 5-min loop; Claude Code occasionally dropped 15-min loops; ChatGPT's monitor loop repeatedly paused itself for 12+ hours with no logs.
- Upgrading from the $20 Plus plan to the $100 Pro plan made ChatGPT seemingly worse, prompting the author to question how this becomes a $1-2T business.
- Takeaway: completely FOSS software and models worked near flawlessly while paid stacks failed, validating the author's investment in local hardware (M2 Ultra 192GB, M3 Ultra 512GB).
More from coding & agent
- OPAL: a reactive optimizer that watches your training and tunes it for you — vincesitzmann · 2026-10-22
- Node.js ships security fixes across 22.x/24.x/26.x, worst bug allows RCE — dotey · 2026-10-22
- Research Progress Graph: modeling authentic research trajectories, 35 agents found incomplete — ZhiyuChen4 · 2026-10-22
- Longshot Studio 0.15 adds lip sync and face refinement, free on your own ComfyUI — R34vspec · 2026-10-22
- Dev builds Scratch-style MCP that feeds coding agents one task at a time to stop plan drift — NoTokenDev · 2026-10-22
- Frontier models spark a comeback of quirky game-style personal homepages — HamelHusain · 2026-10-22