AI Model Explorer: LLM Pricing & Comparison

If you prioritize generation speed and OpenAI SDK drop-in compatibility, Groq Cloud (Llama 3.3 70B at 250+ tokens/sec) is the best starting choice. If you require large context windows and multimodal inputs (PDFs, video, audio), Google AI Studio (Gemini 2.0 Flash with 1M context and 1,500 free daily requests) is the top recommendation. If you want to experiment across dozens of models without changing client code, OpenRouter's free routes provide the highest flexibility.

An OpenAI-compatible API adheres to the standard schema of OpenAI's /chat/completions endpoint. This means you can install the official `openai` library in Python or TypeScript, configure the `base_url` to point to Groq, OpenRouter, or Mistral, and keep the exact same function calling, streaming, and prompt syntax without rewriting your application code.

Free tiers are built primarily for development, sandbox experimentation, and low-volume side projects. Because free tiers typically enforce rate limits between 15 to 30 requests per minute and occasionally permit provider data logging for model training, production systems handling sustained customer queries should budget for pay-as-you-go paid endpoints.

Open Weights models (like Meta's Llama 3.3, Mistral 7B, and DeepSeek R1) release their mathematical weights publicly, meaning you can self-host them locally on your own GPUs (e.g. using Ollama or vLLM) or use hosted inference providers. Closed Source models (like OpenAI's GPT-4o and Anthropic's Claude 3.5 Sonnet) can only be accessed via the creator's proprietary cloud endpoints.

Our directory is audited monthly against official documentation from Google AI Studio, Groq, OpenRouter, GitHub Models, OpenAI, Anthropic, Mistral, and DeepSeek. Last verification cycle: November 2026.