Hugging Face shows ML-Intern: an agent trains six models in a few days for about $103 in total
A machine learning agent that asks for a budget, runs small tests and then trains and publishes, now available in HuggingChat.
// Key points
- ML-Intern plans the work, asks for a budget before spending, runs small tests before full jobs, then trains, evaluates and publishes models on Hugging Face hardware.
- Each task starts with a zero-dollar budget and needs your permission before paid jobs; you use it by turning on ML-intern mode in HuggingChat.
- In one example, Citrus Doctor fine-tuned Qwen3.5-2B for citrus plant problems, raising accuracy on 335 test photos from 14.9% to 52.8% for about $1.90.
- The author built six projects in a few days for about $103 in total; the models, datasets and Spaces are public under the ML-Intern-lab organization.
Builder's takeStart at zero budget and ask before spending is a design every agent that touches money should copy. For small teams, a specialized small model now costs a few dollars per experiment; I'd try it on fixed tasks such as content moderation or tagging first, then compare cost against calling a large model.
// Background · from #AI agents
Full timeline →- Oct 10 Google open-sources AQuA, a quality agent that samples production sessions to diagnose a live agent, at $3.76 for a 32-session sweep
- Oct 8 Local sandboxing for GitHub Copilot is generally available, restricting agent commands’ file, network and credential access with enforceable enterprise policies
- Oct 8 Microsoft makes MXC agent containers generally available on Windows as NVIDIA opens RTX Spark laptop preorders with 1 petaflop of FP4 and up to 128GB of unified memory