Free LLM API tiers useful when prototyping Streamlit apps (Oct 2026 caps)

Random
1

When I prototype Streamlit apps that call LLMs, the free API tiers change often enough that a hard-coded “unlimited” assumption breaks. A few rounded caps I rechecked recently (Oct 2026):

  • Google AI Studio / Gemini free tier: generous daily request caps for Gemini models (treat as changing — check the official rate-limit page).
  • Groq free tier: high tokens/min on open models, but daily caps apply and reset windows matter.
  • OpenRouter free models: useful for routing, but the free model list and RPM move frequently.
  • Cloudflare Workers AI free: limited neurons/day — fine for demos, not load tests.

I keep a public catalog of free AI offers with rounded limits and a check date on each card: https://nopaywall.io

Disclosure: I’m a co-founder of nopaywall.

2

Those free-tier caps match what I’ve been hitting on Streamlit prototypes too. Once Gemini/Groq/OpenRouter free windows are gone for the day, I usually drop the app onto a Haiku-class model instead of keeping a Sonnet/Opus key in .env for every refresh.

On Anthropic’s public list that’s about $0.10 per 1M input / $0.50 per 1M output for Haiku 5.5, versus roughly $2/$10 for Sonnet 5.5. For a Streamlit demo that re-runs the chain on each widget tweak, that gap is the difference between “fine to leave open” and “watch the bill”. I keep a hand-checked table of those rates here: Claude Haiku 5.5 API Pricing 2026 — Cost per 1M Tokens · AI//COST (I help run that site).