How to Compare Free LLM APIs — What Matters

Not all specs matter equally. Here's what to focus on when comparing free models:

• Context window — Most important for coding tools (Claude Code, Cursor). 128K is the practical minimum for processing a real codebase. For long documents or multi-file refactors, look for 256K+.
• Rate limit (RPM) — 30+ RPM is comfortable for solo developers (one request every 2 seconds). Below 10 RPM, you'll feel the throttle. Check both RPM and daily caps (RPD).
• OpenAI compatibility — If the model's API is OpenAI-compatible, it works with virtually every AI tool (Cursor, Codex, OpenCode, LangChain). Non-compatible APIs need custom integration.
• Tool calling support — Essential for AI coding agents. If the model can't call tools, it can generate code but can't run commands or search your codebase.
• Credit card requirement — Some "free" tiers still require card verification. If you want zero friction, filter for "No Card" models on the models page.
• Max output tokens — For coding, 8K output is the minimum (one full source file). 16K+ lets the model generate entire modules. Reasoning models need extra output for chain-of-thought.

After comparing specs, test your top picks in the Playground with a real API key — latency and output quality can't be judged from numbers alone.