AI grifters would have you believe Apple Silicon is suitable for agentic tasks.
Here's M3 Ultra (512GB) running DwarfStar DeepSeek V4 Flash, taking 1 minute to respond to "hi" on an empty chat. MLX performance is even worse than this. But sure, Apple is great for local AI 😆
Hermes Studio perfects FPS using DeepSeek-V4-Flash!
Can Qwen3.8-27B and Bionic surpass this? ⚔️
Agent OS model guide: which AI to use for what job.
Plus one weird trick from the creator of Claude Code. The stack that works right now: → Orchestrator: Claude Fable 5. The best brain to run the whole system. → Agent work: DeepSeek V4 Pro. Very agentic. Very affordable.
DSH-Code is a terminal coding interface for DeepSeek Harness (dsh), optimized for CLI/TUI workflows.
Repo: UNLINEARITY/dsh-code This video shows six animations triggered when switching models or reasoning effort with the official API.
You have reached the end of the archive
All of deepseek