we_are_coded.by CODE · The world, decoded
БГ
Read

In one window, Meta and GitHub close free AI paths - open gets more expensive

MetaControl

Meta is switching off the Llama API Public Preview on July 6; GitHub announced the full retirement of GitHub Models from July 30. The models stay available, but the convenient free doors to them are closing.

In short
  • Meta is switching off the Llama API Public Preview on July 6, 2026; requests return a sunset response, the models stay available for download.
  • GitHub announced on July 1 the full retirement of GitHub Models from July 30, 2026.
  • Read: free hosted AI paths are closing - keep a backup provider or self-host, don't tie yourself to someone else's free plan.
Checked on6 July 2026Responsible editorTsvetelin IvanovHow we workMethod · Corrections

Two quiet announcements landed in the same window, and they point the same way. Meta is killing its free Llama API, GitHub is fully closing GitHub Models (GitHub's service for trying out AI models). Separately - nothing. Together they say something else.

The facts: Meta retires the Llama API Public Preview on July 6, 2026 - after that date, requests return a sunset response (a message that the service is shutting down) with redirect instructions. The Llama models themselves stay free to download, and Meta points developers to third-party providers running Llama. Separately, GitHub announced on July 1 that it's fully retiring GitHub Models on July 30, 2026. Primary sources: llama.developer.meta.com and github.blog.

I've seen this move from the inside: it's not a hit on open models - the weights are out there, anyone can download them. It's a hit on the convenient free door. Hosted free inference (an AI generating a response) was the way to try something without a bill and without a server. Now the trial either costs money or runs through someone else's provider.

A product hanging off a single free hosted API is a risk dressed up as savings. Keep a backup provider, or host the models yourself. Tie yourself to someone else's free plan, and you're tied to their calendar - they decide when it ends, not you.

The visual is generated code art. No third-party images.
Follow usFacebookLinkedIn
Official primary sources
→Meta - Llama API (upcoming changes / sunset)→GitHub Changelog - GitHub Models retirement
Original: https://wearecoded.com/en/articles/closing-free-ai-paths-meta-github.html
ShareFacebookXLinkedInTelegramWhatsApp

More on this

Control
Frontier
Hugging Face27 July 2026Models

The biggest open model is now downloadable: Moonshot releases the weights for Kimi K3

On 27 July Moonshot AI uploaded the full weights for Kimi K3 - 2.8 trillion parameters, the largest openly available model to date. We wrote about its debut ten days ago. Now anyone can download it and run it on their own machine. In theory. In practice the weights weigh in at around 1.5 terabytes and need hardware that few people have.

Read →
Robotics
NVIDIA20 July 2026

NVIDIA released an open model that makes the robot think on the device itself

On 20 July NVIDIA released Cosmos 3 Edge - an open model with about 4 billion parameters that understands its surroundings and decides the robot's next move on its own, right on the device, with no cloud. It runs in real time on their chip, Jetson. For robotics that means less latency and less dependence on an outside connection.

Read →
Frontier
Thinking Machines Lab15 July 2026open models

Thinking Machines open-sourced a frontier-class model

On 15 July the lab led by Mira Murati - OpenAI's former CTO - showed Inkling, its first public model, and released it with open weights. It's not a small demo model: a large mixture-of-experts class, multimodal, with a long context window, that anyone can download and run themselves.

Read →
← Back to all news