DeepSeek's super-agent runs 1.9x faster but burns 3.8x the tokens
AIFlow_ML · x · 2026-11-03
On Nov 1, DeepSeek Harness published a long design article, "Exploring Super-agent Design: Towards a More Autonomous Agent System," reposted by the official account and reaching 1.1M views. The design targets an always-on agent, with measured 1.9x speedup at the cost of 3.8x token consumption.
The code is open-sourced as dsh-superagent-prototype on GitHub (MIT, released Oct 31), explicitly labeled a research prototype, not an official product. It hit 7,637 stars and 775 forks by Nov 2. The author, a one-person AI company builder, breaks down what the test measured and which design ideas are worth stealing.
More from coding & agent
- Good prompts are about articulating taste, not writing specs, argues designer — dreamwieber · 2026-11-03
- Two gaps every AI agent team must understand: requirement gap and model gap — bigdata · 2026-11-03
- Hermes Agent + MeshyAI turns conversational prompts into 3D game characters — egeberkina · 2026-11-03
- First hands-on: Meta's Muse Wonder 1.0 vs DeepSeek V4.1 Flash in Codex CLI — Richidm · 2026-11-03
- "Claude Code has been terrible today" — developer smells a new Opus launch coming — hermanschutte · 2026-11-03
- One command audits every skill loaded by Claude Code, Codex, Cursor and more — 236 found — UnicornAdoption · 2026-11-03