What Is Apertus and What's New in Apertus 1.5 on Public AI Inference Utility?

Updated

Apertus is an open-source language model released by a public institution, and Apertus 1.5 is its latest version, now live on Public AI Inference Utility. According to Public AI's site, the 1.5 release adds three things over the original: image understanding, a context window 4× longer than before, and better tool use — while remaining fully open. If you want to try a sovereign, publicly built model rather than a commercial one, this is the version currently offered on the platform.

What Apertus is

Public AI describes Apertus as "the most powerful open-source language model ever released by a public institution," launched by Imanol's team on September 2, 2025. Two things follow from that description:

  • It's open source. The model weights and code are openly available, not gated behind a proprietary API.
  • It comes from a public institution, not a private frontier lab — which is the "sovereign AI" angle Public AI emphasizes.

Public AI Inference Utility itself is a nonprofit, open-source service whose stated goal is to make public and sovereign AI models more accessible. Apertus is one of the models it hosts.

What's new in Apertus 1.5

The site's release note (dated July 24, 2026) lists three concrete changes:

Capability What changed in 1.5
Image understanding New — the model can now process images, not just text
Context window 4× longer than the previous version
Tool use Improved

The note also stresses that 1.5 is "still fully open," so the upgrade doesn't come with new access restrictions.

One caveat: the site's FAQ asks "how big the context window actually is" as an open question, so the exact token count isn't stated in the material here — only that it's 4× the earlier version. If you need the precise number, check the model card or technical documentation rather than assuming a figure.

How to try it

The homepage's primary call to action is "Try Apertus," and there's a LOG IN entry point in the navigation. So the practical path is:

  1. Go to publicai.co.
  2. Select "Try Apertus" (or log in first if the interface requires it).
  3. Use the model through the utility's interface.

The site doesn't state pricing or whether login is required for every feature, so don't assume free, unlimited, or anonymous access — verify on the page itself.

Who builds and supports it

Apertus isn't a solo effort. Public AI's partner stories mention:

  • AWS Zurich and Intel helped deploy public AI "at national scale" and built the layer that now hosts EuroLLM, ALIA, and other models.
  • Hugging Face — Public AI launched as an official inference provider there (September 16, 2025).
  • Public AI also invites model and inference partners to apply, so the roster of hosted models can grow.

Common confusions to avoid

The FAQ directly addresses a few mix-ups worth knowing before you dig in:

  • Public AI didn't necessarily build Apertus from scratch on top of Llama, Mistral, or DeepSeek — the FAQ poses this as a question, so treat assumptions about its base architecture as unverified until you read the technical docs.
  • publicai.co is not publicai.io or publicai.com. The site explicitly asks whether it has "any link with publicai.io and their crypto coin or publicai.com" — a signal that those are separate entities. Don't send funds or credentials to a similarly named domain.
  • Data handling and privacy are covered in the FAQ ("Where is my data processed and stored?"), so read those answers if you plan to send sensitive input.

When Apertus 1.5 is the right choice

Pick Apertus 1.5 if you want an openly licensed model with image input, long-context handling, and tool-calling, hosted by a nonprofit rather than a commercial vendor. Skip it if you need a guaranteed context size, published pricing, or a specific compliance guarantee — none of those are confirmed in the material here, so confirm them at the source first.