DeepSeek super-agent runs 1.9x faster but burns 3.8x tokens, delegation prompt makes the difference
AIFlow_ML · x · 2026-11-03
A breakdown of a DeepSeek super-agent benchmark article: the super-agent finishes tasks 1.9x faster while using 3.8x the tokens, and the coding step in the test was just a time-burning tool call. The pivotal table (section 7.1) shows a neutral prompt led to 0 delegations in 5 runs, while a delegate-first prompt triggered delegation in 4 of 5 runs with a median of 13.8 min vs 12.4 min for the super-agent. Takeaway: much of the speedup is behavioral willingness to delegate, and the delegation prompt style is what's worth copying.
Related event: DeepSeek Super-Agent Is 1.9x Faster but Burns 3.8x Tokens(4 posts)→
More from coding & agent
- Matt Shumer opens his agent for SMS trials, giving $150 in free credits to new users — mattshumer_ · 2026-11-04
- Model Vendors Post-Train for Tool Search, Squeezing Out MCP-Layer Workarounds — pvncher · 2026-11-04
- Engineer builds agent that clears years-old code backlog without any mandate — BradPorter_ · 2026-11-04
- Sentry CEO: AI Code Verification Is About Design Correctness, Not Compilation — trq212 · 2026-11-04
- AnnotateKit launches: agent-readable in-app feedback for iOS that drives surgical AI code edits — julianweisser · 2026-11-04
- AIOps platform cuts agent token costs 30%+ with Pydantic's open-source sandbox Monty — samuelcolvin · 2026-11-04