DEV Community

TonkaToyXL
TonkaToyXL

Posted on

11 free AI APIs you can use today (with the actual limits)

#ai #webdev #api #beginners

Most "free AI API" lists repeat old numbers. This one uses the vendors' own pricing and docs pages, checked on 2026-10-14. Each entry gives the limit as the vendor states it, the catch, and the source link so you can check it yourself. Free tiers change often, so read the source before you build on one.

The list comes from Free Tier Finder, a directory of free tiers for 134 AI and SaaS tools. Its data is open under CC BY 4.0 at TonkaToyXL/free-tier-finder-data.

LLM inference

1. Google AI Studio (Gemini API)

  • Free: free input and output tokens for some Gemini models, plus AI Studio access.
  • Catch: rate limits per model. Free-tier content is used to improve Google products, and the newest models are not on the free tier.
  • Source: https://ai.google.dev/pricing

2. Groq

  • Free: a free API plan for open models such as gpt-oss-120b, Qwen and Whisper.
  • Limits: per model. For example, 30 requests/min and 1,000 requests/day on gpt-oss-120b; some models are lower.
  • Source: https://console.groq.com/docs/rate-limits

3. OpenRouter

  • Free: chat and API access to 25+ free models from 4 free providers, with activity logs.
  • Catch: no prompt caching, budgets or management keys on the Free plan.
  • Source: https://openrouter.ai/pricing

4. Cloudflare Workers AI

5. Cohere

  • Free: API calls with a Trial API key.
  • Catch: trial keys are rate limited and may not be used for production or commercial purposes.
  • Source: https://cohere.com/pricing

6. NVIDIA NIM

  • Free: serverless API endpoints for open models (for example Nemotron, DeepSeek, Kimi) for prototyping, plus self-hosting NIM on up to 16 GPUs for development.
  • Catch: development, testing and research only. Production needs an NVIDIA AI Enterprise license.
  • Source: https://docs.api.nvidia.com/nim/docs/product

7. Hugging Face

  • Free: Hub hosting, Spaces on free CPU Basic (2 vCPU, 16 GB) and ZeroGPU hardware, and small monthly inference credits.
  • Catch: the inference credits are small, so treat this as a place to prototype.
  • Source: https://huggingface.co/pricing

Search for AI agents

8. Tavily

9. Exa

  • Free: $10 in credits every month plus a $10 onboarding bonus, all endpoints and an MCP server. No card needed, per the pricing page.
  • Catch: credits are spent at standard rates (Search from $4 per 1,000 requests), and Free is capped at 10 queries per second.
  • Source: https://exa.ai/pricing

Speech

10. AssemblyAI

  • Free: $50 in credits on signup for speech-to-text, streaming and the LLM Gateway. No card needed, per the pricing page.
  • Catch: the credits are a one-time amount. Free-tier streaming allows 5 new streams per minute.
  • Source: https://www.assemblyai.com/pricing

11. Cartesia

  • Free: 20K credits a month for text-to-speech and speech-to-text (about 27 minutes of Sonic-3.6 audio), plus $1 a month of voice-agent usage.
  • Catch: no commercial use license or voice cloning on Free. Both start on the $5 Pro plan.
  • Source: https://cartesia.ai/pricing

How to pick

  • Prototyping a chat feature: Google AI Studio, Groq or OpenRouter. Read the rate-limit page for your exact model.
  • Shipping to production for free: check the terms first. Cohere trial keys and NVIDIA NIM are explicitly not for production.
  • Hard daily budget: Cloudflare Workers AI is easy to reason about. It stops at the cap instead of billing you.
  • Agents that need web search: Tavily or Exa.
  • Voice: AssemblyAI for a one-time chunk of credits, Cartesia for a monthly allowance.

About the data

Each entry above was checked against the linked vendor page on 2026-10-14. On Free Tier Finder, each card also has a badge (No card, Card required or Card not stated). If a vendor doesn't say whether a card is needed, the badge says "Card not stated" instead of guessing.

If you spot a number that has changed, open an issue on the data repo and we'll recheck it.

Top comments (0)