INDEPENDENT OPEN-WEIGHT INTELLIGENCE SOURCE FIRST

Open models.Clear decisions.

Explore 811 field-verified AI model records. Understand licenses, compare checkpoints and trace what the original sources actually say.

811 verified models85 curated selections148 source-linked reports
Original midnight violet editorial illustration of a glowing amber adapter prism connected to a large glass neural matrix with fine code-like light trails.
LATEST NEWS · 2026-11-16Tess 4.5 publishes a 327 MB coding adapter for the 594B Inkling 2 Small modelRead the article
CHECKED 2026-11-16 · 811 FIELD-VERIFIED RECORDSHow our evidence works
Evidence standardSource-first verified
Checked2026-11-16
Coverage811 published records

GO FROM UPDATES TO DECISIONS

Know what changed. Choose with evidence.

Three first-party research paths built from our reviewed source records.

NEW · OPEN-WEIGHT AGENT INTELLIGENCE

Build agents on evidence.

Explore 85 Leading and Notable open-weight models by their role in agent systems. Compare reasoning cores, retrieval components and media tools with traceable license, context and hardware evidence.

Explore models for AI agents →
85 curated modelsSix use casesEvidence-first comparisonNo invented agent benchmarks

OPEN WEIGHT INTELLIGENCE

What matters now.

Original source-linked reporting, with a direct route to model evidence. Browse all News →

THREE WAYS TO USE OPEN MODEL WEIGHTS

Find. Verify. Monitor.

Start with a model, inspect its evidence, then follow the changes that matter. No composite rankings or invented compatibility claims.

MODEL-SPECIFIC EVIDENCE ANSWERS

What did we actually observe?

Read a dated source comparison before inferring a model or license change. These are observed examples, not a model ranking.

SOURCE-VERIFIED · NO MODEL RANKINGS

Evidence, at a glance.

Counts from 811 verified model records. Not intelligence, speed or quality scores.

Documented fields

How many records have a verified or classified field?

License declared773 / 811
Commercial-use classified609 / 811
Context verified583 / 811
Explore license evidence →

Source-claim coverage

Current claim-schema refresh progress across the registry.

100.0%811 of 811 records
811 refreshed0 pending

Missing source claims stay unknown, never false.

Inspect source comparisons →

Technical traces

Repository-observed technical evidence, not runtime tests.

Exact weight file lists811 / 811
Runtime evidence signals795 / 811

A runtime signal does not prove successful execution.

Review verification scope →

Bars show field coverage as a share of the verified registry. They do not compare model capabilities. Read the methodology →

TEMPORAL EVIDENCE GRAPH

Search the current ecosystem. Remember how it changed.

Lists are easy to copy. A dated record of what was observed, where it came from and how it changed is harder to recreate.

VERIFIED RECORDS

Explore the evidence, not a leaderboard.

Browse all 811 models →

STAY CONNECTED

Keep open-weight AI within reach.

Choose how to stay up to date. No account or email subscription required.

Follow on X

Discover model news, useful comparisons and important developments, with links back to the evidence.

Follow @openmodelweight

Keep us one tap away

Save the website to your phone’s Home Screen or bookmark it on your computer.

Show save instructions

iPhone / iPad (Safari): Tap Share, then Add to Home Screen.

Android (Chrome): Open the browser menu (⋮), then choose Add to Home screen or Install app, when offered.

Desktop: Bookmark this page with ⌘ + D (Mac) or Ctrl + D (Windows / Linux).

Following individual models? Open your personal watchlist →

PRACTICE GUIDES · REGISTRY-BACKED

Evidence you can put to work.

Explore hardware decisions, source transparency and local coding workflows using our verified model records. Estimates remain estimates, and unknown facts stay unknown.

Explore all practice guides →

SOURCE-FIRST BY DESIGN

Every important claim should have somewhere to point.

Open Model Weights separates repository evidence, publisher documentation and derived values. Unknown stays unknown, and popularity signals never become a quality score.