OPEN MODEL WEIGHTS
The evidence layer
for open-weight AI.
700 field-verified records for exact weight artifacts, licenses, commercial-use conditions, context, formats, lineage and runtime signals — with the source trail kept attached.
INTERROGATE THE REGISTRY.
Start with the question.
The interface stays simple up front. Evidence, provenance and technical detail remain available when you need to go deeper.
Find a model
Search verified records by developer, license, format, context and hardware.
Search the registry →Compare records
Put 2–4 models side by side without composite scores or a declared winner.
Compare models →Trace what changed
Follow repository revisions and field-level evidence changes over time.
Open change history →Query the data layer
Use the versioned JSON API or read-only MCP endpoint in your own tools.
Open API & MCP →EVIDENCE INTELLIGENCE
A registry that gets more useful with time.
Lists are easy to copy. A dated record of what was observed, where it came from and how it changed is harder to recreate.
History that compounds
Observed field snapshots, source revisions and material diffs accumulate instead of being overwritten.
Explore model history →COMPATIBILITYRelationships, not rankings
Connect runtimes, formats, precisions, licenses, base models and memory constraints.
Open compatibility graph →FRESHNESSDaily verification
Repository revisions are checked every day; changed and new sources return to full field verification.
See recent changes →MACHINE LAYERBuilt for humans and agents
Versioned JSON, OpenAPI and a read-only MCP endpoint expose the same evidence as the website.
Connect through MCP →VERIFIED RECORDS
Explore the evidence, not a leaderboard.
NVIDIA Nemotron 3 Nano 30B-A3B
262,144 tokens · nvidia-nemotron-open-model-license
Mistral Small 4 119B A6B
1,048,576 tokens · Apache-2.0
Qwen3-32B
40,960 tokens · Apache-2.0
DeepSeek-R1
163,840 tokens · MIT
gpt-oss-120b
131,072 tokens · Apache-2.0
Gemma 3 27B IT
128,000 tokens · gemma
OPEN WEIGHT INTELLIGENCE
What changed today.
Daily source-first developments tied back to the evidence layer. Open all news →
AstaBrief turns retrieved scientific evidence into cited reports and gives institutions an open-weight option for running the report-writing stage on their own infrastructure.
Read brief →Runtime · 2 Oct 2026llama.cpp adds decision-model inference through a new System One endpointThe local inference stack can now score typed options directly instead of generating free-form text, opening a different deployment path for routing, moderation and agent control.
Read brief →Training infrastructure · 2 Oct 2026Olmo-core 3 opens a training stack designed for trillion-scale mixture-of-experts modelsAi2 redesigned its open training infrastructure around MoE routing, expert parallelism and lower-precision execution, with the next generation of Olmo expected to use the stack.
Read brief →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.