Free LLM APIs that work today
14 providers and 30 models answered our live check on 2026-10-27 (UTC). We run an AI company on free inference, so we probe every model we use each day with a chat request and a tool call; this page is rebuilt from those results: the time is the latest answer's, and "tool calls work" means the model made valid function calls across the last week of probes.
Free tiers change without notice. A model is listed only if it answered within the last three days. Use one account per provider: most terms forbid more.
Most of these speak the OpenAI API, so the official client works with a different base URL:
from openai import OpenAI
client = OpenAI(base_url="<the provider's base URL>", api_key="<your key>")
r = client.chat.completions.create(model="<a model id below>", messages=[{"role": "user", "content": "Hello"}])
print(r.choices[0].message.content)
Cloudflare Workers AI
Sign up · needs: email
Free limits: not published. Notes: says it doesn't train on prompts.
Answered our check (4 of 4 we use): @cf/qwen/qwen3-30b-a3b-fp8 (0.8 s, tool calls work), @cf/openai/gpt-oss-20b (0.4 s, tool calls work), @cf/zai-org/glm-4.7-flash (2.2 s, tool calls work), @cf/google/gemma-4-26b-a4b-it (1.4 s, tool calls work)
Google AI Studio
Sign up · needs: a Google account
Free limits: not published. Notes: free-tier prompts may be used for training or reviewed.
Answered our check (3 of 6 we use): gemini-3.5-flash-lite (5.1 s, tool calls work), gemini-3.6-flash (1.8 s), gemma-4-26b-a4b-it (0.8 s, tool calls work)
Kilo Gateway
Sign up · needs: email or GitHub
Free limits: 3 requests/min. Notes: free-tier prompts may be used for training or reviewed.
OpenAI-compatible base URL: https://api.kilo.ai/api/gateway
Answered our check (3 of 5 we use): nvidia/nemotron-3-ultra-550b-a55b:free (2.3 s, tool calls work), poolside/laguna-s-2.1:free (1.1 s, tool calls work), poolside/laguna-xs-2.1:free (1.7 s, tool calls work)
Groq
Sign up · needs: email + phone verification
Free limits: 30 requests/min, 8,000 tokens/min, 200,000 tokens/day. Notes: says it doesn't train on prompts.
Answered our check (3 of 3 we use): openai/gpt-oss-120b (0.4 s, tool calls work), openai/gpt-oss-20b (0.3 s, tool calls work), qwen/qwen3.8-27b (0.1 s, tool calls work)
Gonka Broker
Sign up · needs: email, no card
Free limits: 5 requests/min. Notes: says it doesn't train on prompts.
OpenAI-compatible base URL: https://proxy.gonkabroker.com/v1
Answered our check (3 of 3 we use): MiniMaxAI/MiniMax-M2.7 (2.9 s, tool calls work), deepseek-ai/DeepSeek-V4-Flash-0731 (40.5 s, tool calls work), zai-org/GLM-5.3-Flash (1.9 s, tool calls work)
Hetzner Inference API
Sign up · needs: email, name, billing address
Free limits: 10 requests/min, 4,000,000 tokens/min. Notes: says it doesn't train on prompts; terms say evaluation or experiments only, not production.
OpenAI-compatible base URL: https://inference.hetzner.com/api/v1
Answered our check (2 of 2 we use): Qwen3.8-27B (0.9 s, tool calls work), Qwen/Qwen3.6-35B-A3B-FP8 (1.1 s, tool calls work)
NVIDIA NIM
Sign up · needs: NVIDIA developer account + phone verification
Free limits: 40 requests/min. Notes: free-tier prompts may be used for training or reviewed; terms say evaluation or experiments only, not production.
Answered our check (2 of 5 we use): z-ai/glm-5.3 (6.4 s, tool calls work), nvidia/nemotron-3-ultra-550b-a55b (1.9 s, tool calls work)
Z.ai
Sign up · needs: email or Google login, no card
Free limits: not published.
Answered our check (2 of 2 we use): glm-4.7-flash (2.1 s, tool calls work), glm-4.5-flash (4.3 s, tool calls work)
Requesty
Sign up · needs: email
Free limits: 5 requests/min.
OpenAI-compatible base URL: https://router.requesty.ai/v1
Answered our check (2 of 2 we use): nvidia/nemotron-3-ultra-550b-a55b (1.3 s, tool calls work), google/gemma-4-31b-it (1.9 s, tool calls work)
Ollama Cloud
Sign up · needs: email, no card, no phone
Free limits: 2 requests/min. Notes: says it doesn't train on prompts.
OpenAI-compatible base URL: https://ollama.com/v1
Answered our check (2 of 2 we use): gpt-oss:20b (0.7 s, tool calls work), gpt-oss:120b (0.4 s, tool calls work)
Typhoon API
Sign up · needs: email, no card
Free limits: 100 requests/min. Notes: free-tier prompts may be used for training or reviewed; terms say evaluation or experiments only, not production.
OpenAI-compatible base URL: https://api.opentyphoon.ai/v1
Answered our check (1 of 1 we use): typhoon-v2.5-30b-a3b-instruct (0.7 s)
AMD Radeon Cloud Token Factory
Sign up · needs: email
Free limits: 5 requests/min.
OpenAI-compatible base URL: https://developer.amd.com.cn/radeon/api/v1
Answered our check (1 of 2 we use): Qwen3.8-Flash-Next (4.2 s, tool calls work)
LLM7.io
Sign up · needs: optional free token
Free limits: not published.
OpenAI-compatible base URL: https://api.llm7.io/v1
Answered our check (1 of 1 we use): default (1.3 s, tool calls work)
Alibaba Cloud Model Studio
Sign up · needs: Alibaba Cloud international account
Free limits: 10 requests/min.
OpenAI-compatible base URL: https://dashscope-intl.aliyuncs.com/compatible-mode/v1
Answered our check (1 of 5 we use): qwen-flash (0.4 s, tool calls work)
Also free, not answering our check right now
Listed with a free tier in our catalog, but none of its models answered our latest check (we may hold no key there, or it was down).