# Local AI Picks

> Local AI Picks lists the latest open-weight AI models by task, with their weight files, memory needs, sources and test status.

## How the numbers are made

- One entry per model family, latest version only.
- A weight option fits a device when it needs at most 85% of the device's memory. Where there is no source or basis for an estimate, memory is shown as Unknown.
- Third-party scores are shown with their source and data date. Scores from different sources are not put in one column or one chart.
- The site gives no ratings, no composite score and no estimated local speed.
- Tests run by this site list every run with its input, settings and environment. Runs are not scored.

## Model pages

Each current model has a page under https://localaipicks.com/models/ with its weight options, memory needs, sources and test status. The directory below lists them all.

## Pages

- [Local AI Picks: open-weight AI models to run on your own hardware](https://localaipicks.com/): The latest open-weight models from 49 labs, grouped by task: language, speech, image, video, music, translation and OCR. Weight files, memory needs, sources and test status.
- [Open-weight AI models: 180 latest releases](https://localaipicks.com/models): Search 180 current open-weight models by task, lab, license and test status. One entry per model family, latest version only.
- [Best local LLM: open-weight language models by memory and score](https://localaipicks.com/best-local-llm): 58 current open-weight language models from 26 labs, with weight files, memory needs where known, sources and test status.
- [Local LLMs for 8 GB of memory: 14 that fit](https://localaipicks.com/best-local-llm/8gb): 14 of 58 current open-weight language models have a listed weight option that needs at most 85% of 8 GB, with the largest option that fits.
- [Local LLMs for 12 GB of memory: 15 that fit](https://localaipicks.com/best-local-llm/12gb): 15 of 58 current open-weight language models have a listed weight option that needs at most 85% of 12 GB, with the largest option that fits.
- [Local LLMs for 16 GB of memory: 16 that fit](https://localaipicks.com/best-local-llm/16gb): 16 of 58 current open-weight language models have a listed weight option that needs at most 85% of 16 GB, with the largest option that fits.
- [Local LLMs for 24 GB of memory: 22 that fit](https://localaipicks.com/best-local-llm/24gb): 22 of 58 current open-weight language models have a listed weight option that needs at most 85% of 24 GB, with the largest option that fits.
- [Local LLMs for 32 GB of memory: 25 that fit](https://localaipicks.com/best-local-llm/32gb): 25 of 58 current open-weight language models have a listed weight option that needs at most 85% of 32 GB, with the largest option that fits.
- [Local LLMs for 48 GB of memory: 28 that fit](https://localaipicks.com/best-local-llm/48gb): 28 of 58 current open-weight language models have a listed weight option that needs at most 85% of 48 GB, with the largest option that fits.
- [Local LLMs for 64 GB of memory: 30 that fit](https://localaipicks.com/best-local-llm/64gb): 30 of 58 current open-weight language models have a listed weight option that needs at most 85% of 64 GB, with the largest option that fits.
- [Best local LLM for coding: open-weight code models and their memory needs](https://localaipicks.com/best-local-llm/coding): 9 current open-weight models for coding from 8 labs, with weight files, memory needs where known, sources and test status.
- [Best local TTS: open-weight text-to-speech models](https://localaipicks.com/best-local-tts): 11 current open-weight models for text to speech from 8 labs, with weight files, memory needs where known, sources and test status.
- [Local image generation: open-weight image models](https://localaipicks.com/local-image-generation): 10 current open-weight models for image generation from 6 labs, with weight files, memory needs where known, sources and test status.
- [Local video generation: open-weight video and 3D models](https://localaipicks.com/local-video-generation): 12 current open-weight models for video and 3d from 8 labs, with weight files, memory needs where known, sources and test status.
- [Local music generation: open-weight music and sound-effect models](https://localaipicks.com/local-music-generation): 5 current open-weight models for music and sound effects from 5 labs, with weight files, memory needs where known, sources and test status.
- [Local translation: open-weight translation models and their memory needs](https://localaipicks.com/local-translation): 5 current open-weight models for translation from 3 labs, with weight files, memory needs where known, sources and test status.
- [Samples: the same prompt on different open-weight models](https://localaipicks.com/samples): 58 audio, image and video outputs from 6 open-weight models on 25 fixed prompts. Every run is shown, with its settings.
- [GPQA Diamond: scores of 14 open-weight models, with memory needs](https://localaipicks.com/benchmarks/gpqa-diamond): 198 graduate-level multiple-choice questions in physics, chemistry and biology. Score is the share answered correctly; guessing gets 25%. Higher is better. Scores for 14 current open-weight models from OpenRouter evals, data from 2026-10-10.
- [τ²-Bench Airline: scores of 14 open-weight models, with memory needs](https://localaipicks.com/benchmarks/tau2-bench-airline): A customer-service simulation: the model plays an airline agent, talks to a simulated customer and calls tools under a written policy. Score is the share of tasks passed. Higher is better. Scores for 14 current open-weight models from OpenRouter evals, data from 2026-10-10.
- [Artificial Analysis Intelligence Index: scores of 15 open-weight models, with memory needs](https://localaipicks.com/benchmarks/artificial-analysis-index): A composite score from Artificial Analysis that combines several evaluations into one number. Higher is better. Its makeup changes between versions, so compare only numbers from the same date. Scores for 15 current open-weight models from Artificial Analysis, data from 2026-10-09.
- [GGUF vs MLX, Q4_K_M, Q8_0: quantization and weight formats for local AI](https://localaipicks.com/quantization): GGUF vs MLX and which to pick, what Q4_K_M, Q8_0 and BF16 mean, and which of 27 current open-weight models publish which quantization levels.
- [Method: sources, memory estimates, scores and tests](https://localaipicks.com/method): Where the model list comes from, how the latest version of each family is chosen, how minimum memory is estimated, how third-party scores are cited, and what this site has tested so far.
- [About](https://localaipicks.com/about): Local AI Picks lists the latest open-weight AI models by task, with their weight files, memory needs, sources and test status.
- [MCP Server for Local AI Models](https://localaipicks.com/docs/mcp): Connect an AI assistant to the Local AI Picks list: find open-weight models by task and memory, read weight sizes, sources and test status. Free, no key.