Agent-built classifier labels 192k docs for $0.70 vs $13-26 with frontier LLMs
vanstriendaniel · x · 2026-09-11
vanstriendaniel shared a "tokenmaxxing" data-labeling workflow: use frontier agents to build small purpose-built classifiers for large-scale data curation, at a fraction of LLM labeling costs.
- Workflow: used Astra agent with SetFit and Hugging Face Jobs, turning 200 agent-labeled examples into a reusable document-purpose classifier, with human review of categories and tricky cases
- Results: classified 191,724 FinePDFs-Edu documents for $0.70 in inference compute, vs an estimated $13–26 with cheap batch LLMs (Gemini 2.5 Flash $13, GPT-5.6 Luna $26); training experiments added $2.90
- tomaarsen endorsed it as how modern ML should be done — agents orchestrating small tools or even training models
The author published the full workflow, mistakes, the reusable model, and a prompt to try yourself.
Related event: Agent-Built Classifier Labels 190K Docs for Just $0.70(2 posts)→
More from coding & agent
- Team-level AI agents: where should shared context and history live? — Al_Grigor · 2026-09-11
- Chaining dependent MCP tool calls: no rollback, duplicate risk — agentrsdg · 2026-09-11
- DeepMind-led paper makes design docs the source of truth, code disposable — SMART regenerates in 1.5-3h for ~$100 — Roger_M_Taylor · 2026-09-11
- MathModelAgent gains traction: auto-solves math modeling and writes a submission-ready paper — jihe520 · 2026-09-11
- alphaXiv open-sources OpenResearch to run parallel research agents with any model — alphaXiv · 2026-09-11
- DeskcommCRM: open-source AI sales CRM with native agents and WhatsApp hits 1k stars — melgarafael · 2026-09-11