Free LLM APIs in 2026
Verified rate limits, no-card endpoints, and OpenAI SDK drop-in configuration for the top free LLM API providers.
Verified Rate Limits & Throughput
| Provider | Models | Rate Limit | Max Speed | Free | Card Required |
|---|---|---|---|---|---|
| Groq | Llama 3.3 70B, Llama 3.1 8B, Gemma 2 9B, Gemma 3 12B | 30 RPM / 14,400 RPD | ~330 tok/s | Yes | No |
| Google AI Studio | Gemini 2.5 Flash, Gemini 2.0 Flash, Gemini 2.5 Pro | 15 RPM / 1M tokens/day | ~150 tok/s | Yes | No |
| Cerebras | Llama 3.3 70B | 1M tokens/day | ~2,100 tok/s | Yes | No |
| OpenRouter | DeepSeek R1 Distill Qwen 32B, Qwen 2.5 Coder 32B, Llama 3.3 70B | 20 RPM | ~45 tok/s | Yes | No |
| Hugging Face | Llama 3.1 8B Instruct, Mistral 7B Instruct | 1,000 req/day | ~30 tok/s | Yes | No |
OpenAI SDK Drop-In Configuration
Swap the baseURL to route any OpenAI SDK-compatible client through alternative providers.
import OpenAI from "openai";
// Groq
const groq = new OpenAI({
apiKey: process.env.GROQ_API_KEY,
baseURL: "https://api.groq.com/openai/v1",
});
// OpenRouter
const openrouter = new OpenAI({
apiKey: process.env.OPENROUTER_API_KEY,
baseURL: "https://openrouter.ai/api/v1",
headers: { "HTTP-Referer": "https://opengpuradar.com", "X-Title": "OpenGPU Radar" },
});
// Google AI Studio
const google = new OpenAI({
apiKey: process.env.GEMINI_API_KEY,
baseURL: "https://generativelanguage.googleapis.com/v1beta",
});Data Privacy on Free Tiers
✓
Groq: Does not train on user prompts. No data retention. LPU architecture processes requests in volatile memory only.
✓
Google AI Studio: Does not use user prompts to train models. Content is retained for 180 days for abuse monitoring only.
✓
Cerebras: Does not train on prompts. All inference runs on dedicated wafer-scale hardware with no shared memory.
⚠
OpenRouter: Free :free models do not train on requests. Paid model APIs may have different policies — check each provider's terms.
⚠
Hugging Face: Serverless inference may log requests for abuse detection. Use Inference Endpoints for private workloads.