For those seeking a free API provider for AI and large language models (LLMs), there are several platforms in 2025 offering free tiers compatible with OpenAI-style endpoints and SDKs that are used in the Polly app. Key players include OpenRouter, Mistral AI, Groq, Google AI Studio, Hugging Face and others.
Several services now provide robust free tiers for developers and researchers:
| Provider | Free Tier Details | Notes |
|---|---|---|
| OpenRouter | 50+ requests/day many :free variants | All free models have :free in their name |
| Mistral AI | Free for prototyping; modest limits | Mistrall small is usually the best model for tasks with Polly |
| Groq | Free access to select models | Llama 3, Mixtral, Gemini. Focus on speed. |
| Google AI Studio | Up to one million tokens/min free (Gemini, Gemma flash) | Gemini 2.5 Flash offers a free tier. |
| Hugging Face | Many hosted models (community shared) with free access | 300+ model selection, open-source choices |
Most free tiers enforce strict rate limits. For example:
- OpenRouter caps free users at 50+ requests per day for
:freemodels, which can be restrictive for the usage of Polly. - Google AI Studio offers up to one million tokens per minute for free, but this is often shared across all users of the free tier, leading to throttling during peak times.
- Groq and Mistral AI impose request-per-minute (RPM) or request-per-day (RPD) limits, which can disrupt workflows if exceeded.
Workaround: Use different free models by different providers for different tasks, or switch between free models once the limit is hit.
Free tiers typically provide access to smaller or older models, which may lack the capabilities of their premium counterparts:
- Mistral AI’s free tier typically restricts users to models like
Mistral Small, which is efficient for most tasks but may struggle with complex creative generation of flashcards and exercises. - Hugging Face’s free hosted models vary in quality, as they are community-contributed. Some may be outdated or lack optimization.
- Groq’s free models (e.g., Llama 3, Mixtral) prioritize speed over depth, making them ideal for low latency but less suitable for tasks like exercise generation.
Free APIs often come with data usage policies that allow providers to:
- Log and analyze inputs/outputs for model improvement (e.g., Mistral AI, Google AI Studio).
- Retain data for undefined periods, raising concerns for sensitive requests.
Note: Review the provider’s Terms of Service and Privacy Policy before using their API.