Malte Ubl: $9 of DeepSeek did a day's worth of work
Malte Ubl says 24 DeepSeek V4.1 Flash sub agents orchestrated by Opus 5.5 cost him about $9, against roughly $2K a day in list-price tokens before. Cheap open weights under a frontier orchestrator is suddenly a real setup, not a thought experiment.
Receipts
Malte Ubl ran 24 DeepSeek sub agents under Opus 5.5 for about $9
Malte Ubl says he put deepseek-v4.1-flash-1017 behind Opus 5.5 as the orchestrator in Claude Code and spent around $9 on DeepSeek, which did 99% of the work. He had 24 sub agents running, each in its own Vercel Sandbox to keep them off his laptop, against a previous average of about $2K a day in list-price tokens.
until now I have spent like $9 on deepseek which does 99% of the work
Theo has T3 Code driving threads across several machines from one MCP server
Theo says multi-machine orchestration is working internally in T3 Code: one thread spanning multiple models, harnesses, apps, operating systems and networks, enabled in the labs section of the latest nightly and, in his words, still very early. A day later he posted spinning up 8 threads on different machines through a single MCP server via Grok Bot.
One thread, multi-model, multi-harness, multi-app, multi-os, multi-machine, multi-network.
Theo says a $700 Linux PC beats his $12,000 Mac at spinning up threads
Theo posted a T3 Code visualization comparing the same work on both machines and said the Mac cost $12,000 and the PC cost $700. He says he is amazed every time at how much slower the Mac is, and is thankful he moved all his code work to Linux. He did not publish the timings in the post text.
Every time I spin up a thread on my Mac I am amazed at how much slower it is than my linux machines.