As recorded at collection: 8 services offering free AI model APIs, including OpenCode Zen

This is a list of services that provide free access to APIs for calling AI models from programs. It includes AI coding tool services that offer access for promotional purposes, services run by telecom carriers for testing, and services operated by participants pooling the computing power of their own computers. You can connect without signing up separately or providing an authentication string, and most services let you connect with the API key left blank or set to any value.

OpenCode Zen's endpoint is https://opencode.ai/zen/v1, and it offers big-pickle and limited-time free models whose names end in free. Kilo Gateway offers models whose names end in free at https://api.kilo.ai/api/gateway; most are from NVIDIA's nemotron family, and recently HY3 has also been free for a limited time. OVHcloud AI Endpoints offers Qwen and GPT-OSS at https://oai.endpoints.kepler.ai.cloud.ovh.net/v1. llm7 offers turbo-tier models codestral-latest and devstral-small-2:24b at https://api.llm7.io/v1, and registering a free key slightly raises the limit. G4F offers Google AI Studio free-tier models as is at https://g4f.space/api/gemini-v1beta. Pollinations offers only openai-fast at https://gen.pollinations.ai/v1. For AI Horde, connect to https://oai.aihorde.net/v1 and set the API key to 0000000000. AI Horde's max_tokens must be set to at least 16, and stop must be provided as an array.

OpenCode Zen has a small per-request text capacity of 200K and uses input content to improve its models. Kilo Gateway also records input content for training and limits requests to 200 per hour per IP address. OVHcloud is limited to 2 requests per minute per model per IP address, making it usable for conversation but unsuitable for code autocompletion. llm7 has limits of 1 request per second, 10 per minute, 60 per hour, and 500,000 tokens per 24 hours. Pollinations processes only 1 concurrent request per IP address and returns an error when tools are called. AI Horde does not support tool calls, is very slow, and has small models, making it barely usable.