01 / verified paths
Verified options and their limits
Each row links to a FreeLLM record and an official provider source. Quotas, regions, account requirements and prices can change, so verify the provider page before relying on an offer.
| Provider / model | What the current record says | Next step |
|---|---|---|
| Groqgpt-oss-120b / gpt-oss-20b · qwen3.8-27b · llama-prompt-guard · whisper-large-v3 | OpenAI-compatible API with per-model RPM, RPD and TPM limits. | Official source ↗ FreeLLM record ↗ |
| CerebrasAll Cerebras-powered hosted models while free credits last | $5 free credits after account creation; current terms belong to the official pricing page. | Official source ↗ FreeLLM record ↗ |
| Hugging Face Inference ProvidersOpen models served via multi-provider inference (chat, embeddings, image and more) | Free tier included; monthly credits and provider availability can change. | Official source ↗ FreeLLM record ↗ |
| SiliconFlowXing4.0-29B · PaddleOCR-VL-1.5 · Hunyuan-MT-7B · BAAI embeddings/reranker · Kwai-Kolors (¥0 listed) | Selected models may be listed at ¥0; login, region and model limits vary. | Official source ↗ FreeLLM record ↗ |
| ModelScopeHosted open models (LLM · MLLM · text-to-image) marked with the API-Inference badge | Registered users can access API-Inference with dynamic rate limits. | Official source ↗ FreeLLM record ↗ |
| LongCat APILongCat-2.5-Preview · LongCat-2.0 | OpenAI-compatible endpoint is documented; a permanent free quota is not confirmed. | Official source ↗ FreeLLM record ↗ |
02 / setup
Quick start with an OpenAI-compatible SDK
Most compatible providers use the OpenAI SDK shape. Replace the base URL and API key, then use the model ID listed in the provider documentation.
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.groq.com/openai/v1",
api_key=os.environ["GROQ_API_KEY"], # export GROQ_API_KEY="your key" first
)
response = client.chat.completions.create(
model="openai/gpt-oss-20b",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)Do not copy a model ID or quota from a different provider. The official source link in each row is the authority.
03 / decision checklist
Choose by integration risk, not just the headline quota
An OpenAI-compatible endpoint is useful when it reduces migration work, but compatibility is not binary. Before switching a prototype or agent, verify the exact behavior your application depends on.
- Authentication and base URL: confirm the provider's documented key format and endpoint rather than copying another provider's configuration.
- Model IDs and feature support: verify streaming, JSON/structured output, tool calling, embeddings and multimodal input separately; providers may support different subsets of the OpenAI API shape.
- Quota semantics: distinguish recurring free tiers from one-time credits, promotional pools and models temporarily listed at zero price.
- Region and reliability: free access may be unavailable in some regions and normally has no production SLA. Keep a fallback provider if the workload matters.
- Data handling: review the provider's privacy, retention and training terms before sending private code, customer data or regulated information.
FreeLLM rule: this page does not imply that OpenAI's official API is free. Every alternative is an independent provider or compatibility layer and is linked to its own official source.
04 / related pages