Mythos is here, it’s time to start tokenmaxxing

TL;DR

• 1

  The subsidy is wild: Theo claims Claude Code's Pro and Max tiers can amount to about $8,000 per month of inference for $200, and he personally logged $4,358 on one laptop plus $1,112 on a Mac Mini in 10 days.

• 2

  Mythos/Fable is temporary on subscriptions: Fable is available on Claude Code Pro and Max only until June 22, after which it leaves those plans because it appears too expensive and compute-heavy to keep subsidized.

• 3

  You can burn a 5-hour window shockingly fast: A single orchestration workflow reviewing three Lakebed PRs hit 1.8 million tokens and 42% of a fresh usage window in under 30 minutes by spinning up parallel sub-agents.

• 4

  Account switching is a real tactic: Theo dual-wields two $200 Claude Code accounts, starts timers early with a throwaway prompt, and uses /login mid-session so ongoing workflows route to a different account before rate limits kill them.

• 5

  The best use isn't cheap code generation, it's management at scale: His most practical workflow has Mythos rank open PRs across repos like T3 Code so he can quickly spot easy merges and high-value work, which surfaced a month-old Magnus PR that got merged within five minutes.

• 6

  Remote agents change behavior as much as model quality does: Running Codex, Claude Code, and Hermes Agent on a Mac Mini via SSH, Tailscale, Discord, and T3 Code remote lets him launch long jobs from his phone, close the laptop, and stop living in "one more prompt" mode.