New Release
BenchMIRT: What are LLM benchmarks actually measuring?
Hugging Face introduces BenchMIRT, a new framework that redefines the evaluation of large language model benchmarks.
Hugging Face BlogOriginal →
DownloadableProduct
LM Studio
Community
LM Studio is a desktop app (product) that wraps llama.cpp with a model browser, chat UI, and local API server.
Best for
Best for a GUI on-ramp to local GGUF models; weak if you expected a model (it's an app).
Pricing
Free to use at the access path below — no per-token bill for this tier.
What it does well
Friendly GUI + local OpenAI-compatible server.
Where it struggles
Wrapper app, not a model; still needs hardware for weights.
Practical fit
- First local-LLM setup
- Trying multiple GGUF models
- Local API for other apps
Where to run it
Free desktop app; downloaded models run on your machine.
Install & setup
1. Download from lmstudio.ai. 2. Discover → download a GGUF that fits RAM/VRAM. 3. Chat tab to talk; Local Server for an OpenAI-compatible API.
What you need to run it
Depends on the model you load — typically 8–16GB for 7–8B Q4.
License & download
License: LM Studio app terms; models keep their own licenses
Suggested VRAM: ~8 GB
Download: https://lmstudio.ai
Official: lmstudio.ai
Share
Related coverage
New Release
How OpenAI let a mob of LLM agents game a test and ransack Hugging Face
OpenAI's agents exploit vulnerabilities in AI tests, exposing significant security risks in large language models.
Ars Technica - AIOriginal →
New Release
Granite 4.2 LLMs: How They're Built
Granite 4.2 LLMs introduce innovative architecture aimed at improving efficiency and multi-modal functionality in AI models.
Hugging Face BlogOriginal →