The N44 Model Library

Find the right AI for you.

Discover open AI models, compare capabilities, and find a starting point for your hardware.

Guided Model Finder

What can your hardware run?

Choose your device and what you want to do. We’ll check a small, source-traceable set of runtime variants.

Your main task

How to find your GPU

Explore models

The model library.

Curated entries · Publisher sync not yet run · Benchmarks require sourced results

25 models found

Google

Gemma 3 1B IT

1B

Small text-only assistant for constrained devices.

General

License

Gemma license

GPU estimate (4-bit)

3 GB+

Hugging Face

SmolLM2 1.7B Instruct

1.7B

Lightweight text assistant for experimenting on modest hardware.

General

License

Apache 2.0

GPU estimate (4-bit)

3 GB+

Qwen

Qwen3 4B

4B

Compact text model for everyday questions and lightweight reasoning.

GeneralReasoning

License

Apache 2.0

GPU estimate (4-bit)

5 GB+

Microsoft

Phi-4 Mini Instruct

3.8B

Small instruction model for constrained local setups.

GeneralReasoning

License

MIT

GPU estimate (4-bit)

5 GB+

Meta

Llama 3.2 3B Instruct

3B

Compact instruction model suitable for smaller local machines.

General

License

Llama 3.2 Community License

GPU estimate (4-bit)

5 GB+

Google

Gemma 3 4B IT

4B

Compact instruction-tuned model with image-and-text capabilities.

GeneralVision

License

Gemma license

GPU estimate (4-bit)

6 GB+

Review the Gemma terms before use.

Qwen

Qwen3 8B

8B

A versatile instruction model with thinking and non-thinking modes.

GeneralCodingReasoning

License

Apache 2.0

GPU estimate (4-bit)

8 GB+

Meta

Llama 3.1 8B Instruct

8B

Widely supported instruction model with a custom community license.

GeneralCoding

License

Llama 3.1 Community License

GPU estimate (4-bit)

8 GB+

Access may require acceptance of license terms.

Qwen

Qwen2.5 Coder 7B Instruct

7B

Focused coding assistant for code generation and explanation.

Coding

License

Apache 2.0

GPU estimate (4-bit)

8 GB+

DeepSeek

DeepSeek R1 Distill Qwen 7B

7B

Distilled reasoning model using a Qwen base.

ReasoningGeneral

License

MIT

GPU estimate (4-bit)

8 GB+

DeepSeek

DeepSeek R1 Distill Llama 8B

8B

Distilled reasoning model built from a Llama-family base.

ReasoningGeneral

License

Llama license terms apply

GPU estimate (4-bit)

8 GB+

Mistral AI

Mistral 7B Instruct v0.3

7B

Popular compact instruction model with broad tooling support.

GeneralCoding

License

Apache 2.0

GPU estimate (4-bit)

8 GB+

Mistral AI

Ministral 8B Instruct 2410

8B

Smaller Mistral-family assistant; check terms for commercial deployment.

General

License

Mistral Research License

GPU estimate (4-bit)

8 GB+

IBM

Granite 3.3 8B Instruct

8B

Business-oriented instruction model for text and tool workflows.

GeneralCodingReasoning

License

Apache 2.0

GPU estimate (4-bit)

8 GB+

Mistral AI

Mistral NeMo Instruct

12B

Multilingual assistant model built for everyday text and code tasks.

GeneralCoding

License

Apache 2.0

GPU estimate (4-bit)

10 GB+

Qwen

Qwen3 14B

14B

Mid-size reasoning and instruction model for capable local hardware.

GeneralCodingReasoning

License

Apache 2.0

GPU estimate (4-bit)

12 GB+

Qwen

Qwen2.5 Coder 14B Instruct

14B

A more capable coding model for local development workflows.

Coding

License

Apache 2.0

GPU estimate (4-bit)

12 GB+

Google

Gemma 3 12B IT

12B

Multimodal assistant with image-and-text support.

GeneralVision

License

Gemma license

GPU estimate (4-bit)

12 GB+

Microsoft

Phi-4

14B

Compact reasoning-focused text model.

GeneralReasoning

License

MIT

GPU estimate (4-bit)

12 GB+

DeepSeek

DeepSeek R1 Distill Qwen 14B

14B

Mid-size distilled reasoning model.

ReasoningGeneral

License

MIT

GPU estimate (4-bit)

12 GB+

OpenAI

gpt-oss-20b

20B

An open-weight reasoning model designed to run locally with supported runtimes.

ReasoningCodingGeneral

License

Apache 2.0

GPU estimate (4-bit)

16 GB+

Check runtime support for its native quantization.

Qwen

Qwen3 32B

32B

Larger dense model for advanced local reasoning workloads.

GeneralCodingReasoning

License

Apache 2.0

GPU estimate (4-bit)

24 GB+

Qwen

Qwen2.5 Coder 32B Instruct

32B

Large coding-oriented model for supported local inference hardware.

Coding

License

Apache 2.0

GPU estimate (4-bit)

24 GB+

Google

Gemma 3 27B IT

27B

Larger multimodal model for capable local servers.

GeneralVision

License

Gemma license

GPU estimate (4-bit)

24 GB+

Meta

Llama 3.3 70B Instruct

70B

Large text instruction model best suited to high-memory systems.

GeneralCoding

License

Llama 3.3 Community License

GPU estimate (4-bit)

48 GB+

Hugging Face downloads are usage/popularity signals, not quality benchmarks. Publisher snapshots are refreshed manually and may become outdated. Memory values are illustrative estimates for selected quantized GPU inference configurations, not minimum requirements or performance scores. CPU offloading, format, context length, and runtime can change fit substantially. Always check the publisher's model card and license.

Benchmarks & rankings

Compare performance with context.

Explore documented benchmark results. Scores are only ranked when they share a benchmark version, evaluation setup and score unit. Missing results stay missing—not zero.

Methodology / benchmark source

Coding problem success under a defined benchmark and evaluation window.

Verified results coming soon

No source-audited, protocol-matched results are published for this benchmark yet. N44 won't generate a leaderboard from estimates or unrelated benchmark scores.

Benchmark comparison

Select up to three entries above. This comparison uses the same evaluation cohort, not an invented aggregate score.

No comparable results selected.

Benchmark scores alone do not describe local inference speed, memory use, safety or real-world suitability. N44 does not currently publish first-party standardized performance measurements.

Compare models.

Select up to three models in the library to compare their basic specs.

Choose models above to start comparing.

New to local AI?

Start with a smaller model, check your available memory, and choose a compatible runtime such as Ollama. Download only from trusted publisher or runtime sources and review licenses before commercial use.

Model formats and runtime compatibility vary. These are external resources; N44 does not host or verify every downloadable file.