AI model finder

Which AI runs on your PC?

Start with the hardware you have. Discover local AI models, filter by task and capabilities, and compare their memory requirements.

Hardware

Your hardware

Model
Filter models1

Model type

What kind of AI model do you want to use?

Capabilities
Additional filters
Workload and memory assumptionsAdjust the parameters to your scenario.

Matching models

No matches for these filters.

Understand. Then decide.

Good decisions need context.

Find your hardware specifications

On Windows, Task Manager shows the GPU and dedicated GPU memory under Performance. On macOS, About This Mac shows the chip and shared memory. On Linux, system information and GPU vendor tools provide these details. Enter the memory actually available to AI after accounting for the operating system and other applications.

Understand the fit categories

Memory likely sufficient means the known, calculable components fit the checked pools. Possible with adjusted settings may require explicit offloading. If even the known memory requirement exceeds capacity, the result says so. Missing components leave the assessment incomplete. These categories are independent of speed and software compatibility.

Which model is best?

First select a task, then a relevant benchmark. Transcription errors, mathematics tests and image evaluations use different metrics and do not produce a meaningful combined score. Higher is better for some tests, while error rates should be lower. Agent benchmarks measure the model together with its agent system and tools.

New models and benchmark context

A model release date and repository creation date are different facts. A new repository may contain an older model variant. Base model scores do not automatically apply to every quantization. Inspect the test version, dataset, measured variant, measurement date and source.

Local AI, licenses and access

Public weights do not automatically permit unrestricted commercial use. The finder displays license identifiers and access restrictions where structured data is available. Undocumented capabilities remain unknown. Strict capability filters include documented matches; unclear candidates can be included explicitly.

Frequently asked questions

Do I need an account?

No. All results are freely accessible. Your requirements stay in this browser unless you explicitly share the configuration.

Does sufficient memory guarantee speed?

No. Speed requires a matching measurement. The tools show memory, compatibility and performance separately.

How are planning values derived?

Published hardware guidance takes priority. Otherwise, the calculator derives a conservative value from checkpoint size or parameter count, precision and task-specific working memory. The result links its basis and remains distinct from a measured benchmark.