By Tobias Mann and Thomas Claburn
Publication Date: 2026-05-02 11:30:00
As model developers enforce more aggressive rate limits, increase prices, or even abandon subscriptions in favor of usage-based pricing, this Vibe-encoded hobby project is becoming significantly more expensive. Fortunately, there are also ways to save money.
In recent weeks, we’ve seen Anthropic try to drop Claude Code from its cheapest plans, while Microsoft skipped testing and moved GitHub Copilot to a usage-only model. The whole debacle got us thinking. Do we even need the top models from Anthropic or OpenAI or can we get away with a smaller local model? Sure, it might be slower, less powerful, and a little more frustrating to work with, but you can’t beat the price of the free version… Assuming you already have the necessary hardware.
Coincidentally, Alibaba recently dropped Qwen3.6-27B, which the cloud and e-commerce giant boasts of offering “flagship coding performance” in a package small enough to run on a 32GB M-series Mac or a 24GB GPU.