GitHub Models Free Tier: Retired July 2026

GitHub Models is retired. As of July 30, 2026 it is no longer available to any customer, including existing customers with active usage. There is nothing to claim. This page explains what the free tier was, how it ended, and where to go instead.

GitHub Models was GitHub's built-in way to try AI models: a playground, a model catalog and an inference API, free within rate limits for anyone with a GitHub account. GitHub closed it to new customers on June 16, 2026 and shut it down for everyone on July 30, 2026. If you relied on it, the practical replacement is another free model API with an OpenAI-compatible endpoint — options are compared below.

What GitHub Models was

GitHub introduced GitHub Models in August 2024, pitching it as a way for developers on GitHub to become "AI engineers" by experimenting with models where their code already lived. The launch catalog included Llama 3.1, GPT-4o, GPT-4o mini, Phi 3 and Mistral Large 2, and the built-in playground let you test prompts and parameters for free. The intended path was to start in the playground, move to code in Codespaces or VS Code, and then take the project to production on Azure AI.

Over time it grew into several pieces:

• Playground — a chat-style interface on github.com to compare models and tune prompts and parameters.
• Model catalog — models from providers such as OpenAI, Meta, Microsoft, Mistral and DeepSeek, browsable inside GitHub.
• Inference API — an endpoint you could call from code using a GitHub token, which made it popular for scripts, GitHub Actions and quick prototypes.
• Prompts and evaluations — tooling for working with prompts and comparing model outputs, which BYOK later extended to GitHub Actions as well.
• Bring your own key (BYOK) — added in June 2025, letting organizations connect their own OpenAI or Azure AI keys so usage was billed by that provider.

How the free tier worked

The free tier was rate-limited rather than credit-based. GitHub's documentation described limits on requests per minute, requests per day, tokens per request and concurrent requests, with different limits for "low" and "high" model tiers and higher limits for users on paid Copilot plans. Free usage was explicitly meant for prototyping and experimentation, not production traffic.

On June 24, 2025, GitHub added a way past those limits: opt-in pay-as-you-go billing through your GitHub account (payable by credit card, PayPal or invoice), organization-level controls over which models members could use, and BYOK. GitHub said model access remained "free to start within our rate limits", that data stayed on GitHub and Azure infrastructure, and that it was not used for training. Paid usage was off by default for organizations and enterprises, with a $0 spending limit until an admin raised the budget.

Retirement timeline

Why it ended

GitHub hasn't published a reason. The June 16 post simply says "We are retiring GitHub Models", and the July posts confirm dates and alternatives without explanation. We haven't found an official statement beyond that, so we won't speculate about one.

What GitHub recommends instead

• Microsoft Foundry (ai.azure.com) — GitHub's suggested home for new and existing projects that need access to a broad model catalog. It is a paid Azure service; free Azure credit routes include Azure for Students and Microsoft for Startups.
• GitHub Copilot — for AI-powered workflows on GitHub itself, with access to a range of models. GitHub's docs stress that GitHub Models was separate from Copilot, so Copilot plans were not affected. See GitHub Copilot pricing and the free Copilot offer for students.

Who was affected, and what to do now

In every case there is no workaround that restores GitHub Models itself: the official posts are explicit that the service is unavailable to all customers, including those with active usage, and the related UI has been removed from GitHub.

Free alternatives for API access

Most people used GitHub Models for one thing: a free API key to call a capable model from a script or prototype. These free tiers on our site cover the same need. All of them offer an OpenAI-compatible endpoint or SDK, so migrating usually means swapping the base URL, key and model name.

If you need the OpenAI models that were in the catalog, they're available directly from OpenAI's paid API or through Microsoft Foundry. For a full walkthrough of getting a no-cost key, see our guide to free AI API keys.

Migrating code that used GitHub Models

1. Find every call site. Search your repositories, scripts and GitHub Actions workflows for the GitHub Models endpoint and for GitHub tokens used as model API keys. Calls made from Actions workflows are easy to miss.
2. Pick a provider per use. Prototype chat → Gemini, Groq or OpenRouter's free models; embeddings → Cloudflare Workers AI or a dedicated embeddings tier such as Jina AI or Voyage AI; production with OpenAI models → OpenAI directly or Microsoft Foundry.
3. Create a key and store it as a secret. In GitHub Actions, add it under the repository's or organization's Actions secrets rather than hard-coding it.
4. Change the base URL, key and model ID. With the OpenAI SDK this is usually a two- or three-line change. Model names differ by provider, so check each provider's model list.
5. Re-check limits and output. Free tiers have different per-minute and per-day limits than GitHub Models had, and the same open model can behave slightly differently across hosts. Re-run any evaluations you kept in your repository.
6. Remove dead configuration. Delete stored BYOK settings and GitHub Models prompt files you no longer need so teammates don't assume the service still works.

# Example: same OpenAI SDK, new provider
import os
from openai import OpenAI

client = OpenAI(
    base_url="https://<new-provider-openai-compatible-url>/v1",
    api_key=os.environ["NEW_PROVIDER_API_KEY"],
)
resp = client.chat.completions.create(
    model="<model id from the new provider>",
    messages=[{"role": "user", "content": "Hello"}],
)

What we don't know

• Reason for retirement — not stated by GitHub.
• Data retention — the retirement posts don't say what happens to stored prompts, evaluations or BYOK settings. If this matters for compliance, ask GitHub Support.
• Final billing — the posts don't cover refunds or how outstanding pay-as-you-go usage is invoiced. Check your account's billing page.
• Return — GitHub hasn't indicated any plan to bring the service back.

Lessons for choosing a free AI API

GitHub Models' shutdown came with about six weeks between the new-customer cutoff and full retirement. That's a reminder that free tiers are a business decision, not a guarantee. Keeping your model provider behind a small wrapper (one function that takes a prompt and returns text), using OpenAI-compatible endpoints, and storing the base URL and model name in configuration rather than code makes the next migration a configuration change instead of a rewrite. We track status changes like this one and date every check on this site.