AIToday
Large Language ModelsAI Coding AssistantsZenn AI/MLPublished: Nov 19, 2026, 16:00 JST

Claude Code subagents cut 50-account TikTok pull to $0.26

Send on LINE
Claude Code subagents cut 50-account TikTok pull to $0.26

3 Key Points

  1. What happened

    The author pulled 18 verified TikTok brand accounts at 38,794 bytes each — about 9,700 tokens — and compressed each to a 382-byte, roughly 95-token summary, cutting 50 accounts to about $0.26 against about $4.45 without splitting.

  2. Why it matters

    The saving comes from keeping bulky raw data out of the main session and passing only a one-line summary forward, so the gap between what is read and what is retained drives the roughly 17-fold cost difference.

Not sure about something? Ask the AI

Questions and answers are published on this page.

Summaries like this, every morning.

Context & Analysis

The write-up sits against Anthropic's own October 22 guide, which argues most agents should reduce subagents and default to a single model, context window and tool loop. The author agrees in principle but keeps a one-level split limited to a "fetch and compress" step, because each TikTok account returns far more data than the comparison table needs. That matches two of the conditions Anthropic's single-agent-loop-kit README lists for keeping subagents: parallel research in fresh contexts, and work that reads far more than the main agent needs later. A separate finding concerns pacing rather than model choice. With a test key capped at 10 requests per minute, six workers produced 20 rate-limit errors out of 56 requests, while four workers sharing a client-side limiter of 9 starts per minute completed all 12 calls with no 429s. The author frames this as a scheduling problem, not a data problem, and notes that the test key's cap is far below the account default of 60 requests per minute, so the worker count should be recalculated against each user's own limit.

FAQ
How much does the EveryInfra API cost per account?
Each of the three calls per account costs $0.000556, so one account costs $0.001668 and 50 accounts cost $0.0834.
Why use the full model ID instead of the haiku alias?
The haiku alias points to Haiku 5.5 on the Anthropic API but still points to Haiku 4.5 on AWS, Bedrock, Google Cloud and Microsoft Foundry, where Haiku 4.5's list price is 10 times Haiku 5.5's.
What happens when the API rate limit is hit?
The API returns a 429 with retry_after of 6 to 39 seconds; all 20 such errors in the test were not charged and succeeded on the first retry after waiting.
Send on LINE

AI news that matters for your work, delivered every morning.

Pick your industry and the AI tools you use, and get news related to your work every day.

Free · 30 seconds with Google · unsubscribe anytimeWhat is AIToday? →

Ask AI

Ask AI anything about this article. The AI reads this article, earlier AIToday articles, and Wikipedia, and cites its sources. Q&As are published on this page for other readers too.

Questions and answers are published on this page.

Related Articles

Next articleHempfield schools hold off on CurvePoint AI gun detectors after false alarms