

LM Studio
#3lm-studio · seit Mai 2023 (v0.1.x); aktuell v0.4.x (v0.4.0 am 28. Januar 2026) · 33× · zuletzt 30. Juni 2026
61
Momentum
LM Studio ist eine Desktop-Applikation von Element Labs, Inc. zum lokalen Ausführen von Open-Source-Large-Language-Models (LLMs) auf der eigenen Hardware – ohne Cloud-Anbindung. Die Software bietet eine grafische Oberfläche, einen integrierten Hugging-Face-Modell-Browser, eine OpenAI-kompatible REST-API sowie einen headless-Daemon (llmster) für Server- und CI-Deployments. Als Inference-Backends werden llama.cpp (GGUF) und Apple MLX genutzt; die App unterstützt macOS, Windows und Linux und ist seit Juli 2025 für private und kommerzielle Nutzung kostenlos.
Momentum-Verlauf
04.04.03.07.
Features
| Deployment (Self-host/Cloud) | Lokal (Desktop-GUI oder headless via llmster-Daemon); Self-host auf Linux-Servern, Cloud-Instanzen und CI; kein Cloud-Betrieb durch LM Studio selbst |
| Durchsatz/Latenz | MLX auf Apple Silicon ~30–50 % schneller als llama.cpp/Metal; Flash Attention für CUDA/Vulkan/Metal; Continuous Batching für parallele Requests |
| Lizenz | Proprietäre App-Lizenz (Element Labs, Inc.); CLI (lms) unter MIT-Lizenz; kostenlose Nutzung privat & kommerziell |
| Plattform | macOS (Apple Silicon), Windows (x64/ARM64), Linux (x64, ARM); iOS/iPadOS via Locally-App |
| Preis | Kostenlos (privat & kommerziell); Enterprise-Plan für SSO, Model-Gating, private Kollaboration auf Anfrage |
| Protokoll-Kompatibilität | OpenAI-kompatible REST-API (/v1/chat/completions, /v1/completions, /v1/embeddings, /v1/models, /v1/responses) auf localhost:1234; Anthropic-kompatibles /v1/messages-Endpoint; MCP (Model Context Protocol) Host; Python- & TypeScript-SDK |
| Release-Datum | Mai 2023 (v0.1.x); v0.4.0 am 28. Januar 2026 |
| Unterstützte Modelle/Provider | GGUF (llama.cpp) & MLX-Modelle von Hugging Face; u. a. Llama, Qwen, Mistral, DeepSeek, Gemma, Phi, gpt-oss; integrierter Modell-Katalog (model.yaml) |