As of 2026-05-16

As of 2026-05-16

This article is a methodology-first snapshot of the open-weight landscape. It deliberately keeps specific model names short to what we can verify against primary sources at this date, with links so the reader can confirm. Older snapshots live in the archive at their original slugs.

Methodology

Every claim below is tagged according to the labeling policy in how-to-read-llm-release-rumors. Three labels, no untagged claims, every claim with a primary-source URL.

We name families here (the Llama line, the Qwen line, the Gemma line, etc.) and link to each lab's official model card index. For the exact "what is the latest weight you can pull right now" answer, the Hugging Face org page for each lab is the most reliable canonical source — newer than any article, including this one.

What has shipped (Confirmed)

For "shipped" we require a public model card with downloadable weights from the lab itself. Anything that lacks a model card with weights is not Confirmed-shipped, no matter how loudly it has been previewed.

Verified at this snapshot date:

  • Gemma 4 family (Google DeepMind), open-weight release first published 2026‑04‑03. Includes 2B, 9B, and 27B sizes in both base and instruction-tuned variants. Primary: Gemma overview and the Hugging Face org google/ (e.g. google/gemma-4-27b-it).

For every other family in scope (Llama, Qwen, Mistral, DeepSeek, GLM, Phi), the canonical "what is the latest open weight" answer is the lab's own Hugging Face organization page:

We point at the org pages instead of naming specific versions because version names move faster than this article should. If a snapshot article in the Local Models cluster is dated within the past 30 days, prefer it over this list. Otherwise check the org pages.

What is announced but not yet shipped

A model is in this bucket when the lab itself has publicly committed to a future open-weight release but the weights are not yet downloadable. The qualifying signals:

  • An official lab blog or press post announcing a forthcoming open-weight model line.
  • A keynote or paper from the lab where they say "weights will be released."
  • A Hugging Face placeholder repo or model card stub published by the official org with a stated ship window.

When weights land, the claim moves out of this section into Confirmed-shipped above, and the dated article describing the head-to-head moves into the Bench cluster.

We do not list specifics here because they age in weeks. The current state of "announced but not shipped" lives at each lab's official posts.

What is strong-signal

Strong signal means multiple independent reputable reports, or one report backed by an artifact (config file, server endpoint, model card stub without weights). The full list of currently strong-signal claims for individual model families lives in the per-family articles:

This overview article does not enumerate strong-signal claims by name, because the durable thing to say here is the methodology, not the current list.

What is pure speculation

Speculation in this tracker is cadence-based, structural, or inferential. It is labeled as such and is not used as the basis for predictions of capability or release date. Specific cadence claims (e.g. "Llama refreshes every N months") are house estimates rather than verified statistics; when we cite a number we attach the underlying release log used to compute it.

The honest read of 2026 open-weight cadence: the major labs ship faster than any monthly article can keep up with, and most "open-weight release" predictions older than a few weeks are stale by the time you read them. Treat this article's specifics with that in mind, and lean on the primary-source links for current state.

What this tracker is not

Not a leaderboard. Not a buying guide. Not a venue for unsourced screenshots. The Bench cluster handles head-to-heads with disclosed methodology. The Local Models cluster handles which model to actually run on your hardware. The Radar handles "what is coming, with what confidence" — and where the confidence is low, this article says so plainly rather than inventing detail.

Edit log

  • 2026-05-16 — initial publication of this snapshot. After an internal fact-check pass (Perplexity Sonar Pro against the article's claims), most enumerations of specific recently-shipped open-weight models were removed because they could not be verified against primary sources. The article now points readers at lab org pages for live state. Verified specifics: Gemma 4 family release on 2026-04-03 (source).