EVIDENCE-FIRST · NO MODEL RANKINGS
Find models that fit your constraints.
Filter verified records by documented license classification, context window and weight-only memory budget. Every result links to the original evidence – not an invented intelligence score.
Define your requirements
Only activate conditions you can justify for your use case.
Important: Memory estimates do not prove a runnable quantized artifact, fit in available GPU memory, latency or quality. License classifications are not legal approval. Unknown data is never treated as “no”.
SOURCE-BACKED RESULTS
Matching records —
Loading verified registry…
Alphabetical, not ranked by quality, safety, speed or price. Exclusions mean the active fields were not evidenced or the numeric criterion was not met.
Loading model evidence…
How to interpret a model-fit shortlist.
Evidence, not recommendations
All numeric and license filters operate only on explicitly observed or derived model fields. No subjective score, benchmark or blanket licensing assurance is implied.
Weight-only estimates
BF16, FP8 and INT4 estimates derive from the documented total parameter count. KV cache, activations, serving overhead and the availability of quantized weight files are separate questions.
Inspect the source
Review each result’s methodology, original repository, license conditions and observed history. Save candidates in the browser-local shortlist.
Common questions
Does a memory-budget match guarantee that a model will run?
No. Weight-only decimal-GB estimates exclude KV cache, runtime overhead, activations and sharding. They do not prove that matching precision artifacts exist.
Does “commercial use allowed” mean legal clearance?
No. This is a source-based classification, not legal advice. Check the full license and all conditions for your specific usage.
Does this tool rank model quality or inference speed?
No. This finder filters documented characteristics and derived weight-memory numbers. Results are alphabetical and no runtime benchmark is inferred.
How does this tool handle missing information?
Unknown remains unknown. When you require a field such as a context threshold, records lacking that field are excluded from this particular filtered view; exclusion does not mean the model lacks the capability.