Capability strength
Breadth and documented performance across real workloads carry most weight in the overall order.
Open model directory
Open models.
Clear choices.
A researched directory of dependable open-source and open-weight AI models for language, reasoning, code, vision, speech, image generation and retrieval.
43CURATED MODELS
7CAPABILITY AREAS
Open source and open weight
Primary model sources
Last researched 27 Jul 2026
Model directory
Choose for the
system you have.
This is not a single-score leaderboard. The order weighs documented capability at 60 percent and current adoption at 40 percent, with license clarity and deployability as tie-breakers. The right model still depends on the job, hardware, latency target and level of operational control required.
No matching models
Try a broader capability, another deployment scale or a different search term.
Selection method
Dependable means
more than capable.
We rank models using documented capability, current adoption, clear official sources and a realistic path from evaluation to production.
Adoption signal
Downloads, active tooling and production use show whether a model has earned real operational trust.
License clarity
Open-source and open-weight releases are labeled separately so usage obligations remain visible.
Deployability
Memory, context, throughput and infrastructure requirements must align with the real operating environment.
Read the label
Open weights are
not always open source.
Weights, data, code, recipes and intermediate artifacts
The release provides enough of the development process for deeper inspection and reproducibility, not only the final downloadable parameters.
Downloadable parameters with stated license terms
Weights can be inspected or self-hosted. Some use permissive licenses such as Apache 2.0 or MIT, while others add attribution, acceptable-use or commercial conditions.
Always review the current official license and acceptable-use terms before commercial deployment. This directory is technical guidance, not legal advice.
Putting models to work
From model choice
to working system.
ADOR.IS designs AI-enabled products, retrieval systems, model-serving infrastructure, evaluation pipelines and dependable operational software.
[email protected] ↗