Funktion · 2026-05-12

Open-Weight KI-Modelle

Modelle, die trainierte Gewichte unter einer Lizenz zum Download bereitstellen.

Was ist das?

Open-Weight-Modelle veröffentlichen ihre Gewichte (oft Apache 2.0, MIT oder eine Forschungslizenz).
Hinweis: Open weights ≠ Open Source — Trainingsdaten und Code sind meist nicht offen.

Warum wichtig

Self-Hosting auf eigenen GPUs, Fine-Tuning auf privaten Daten, Offline-Betrieb, regulierte Umgebungen.
Die Tabelle zeigt den günstigsten API-Host — mit eigener Hardware kann dasselbe Modell ohne API-Kosten laufen.

414 Modelle mit dieser Funktion

Modell	Anbieter	Eingabe / 1M	Ausgabe / 1M	Kontext	Hoster
Whisper Large v3	scaleway	$0.003	Unknown	Unknown	1
Voxtral Small 24B 2507	Mistral	$0.002	$0.002	32K	3
KB Whisper	evroc	$0.002	$0.002	448	1
Multi-QA-mpnet-base-dot-v1	digitalocean	$0.009	Unknown	512	1
All-MiniLM-L6-v2	digitalocean	$0.009	Unknown	256	1
BGE Reranker v2 M3	digitalocean	$0.010	Unknown	8K	1
Llama 3.2 1B Instruct	Meta	$0.010	$0.010	16K	5
Llama Prompt Guard 2 22M	Meta	$0.010	$0.010	512	2
Llama Prompt Guard 2 86M	Meta	$0.010	$0.010	512	2
E5 Large v2	digitalocean	$0.020	Unknown	512	1
BGE M3	digitalocean	$0.020	Unknown	8K	1
dots.ocr	chutes	$0.010	$0.011	131K	1
Gemma 3 4B	Google	$0.010	$0.027	33K	7
PaddleOCR-VL	novita-ai	$0.020	$0.020	16K	1
Meta-Llama-3.1-8B-Instruct	Meta	$0.020	$0.030	128K	25
Llama 3.1 8B Turbo	Meta	$0.020	$0.030	131K	2
Mistral Nemo Instruct 2407	Mistral	$0.020	$0.040	128K	8
Gemma 3n 4B	Google	$0.020	$0.040	8K	5
Hermes 4 14B	chutes	$0.014	$0.054	41K	1
Meta-Llama-3-8B-Instruct	Meta	$0.030	$0.040	8K	8
Llama Guard 3 8B	Meta	$0.020	$0.060	8K	3
Ministral 3B	azure-cognitive-services	$0.040	$0.040	128K	1
Stable Diffusion 3.5 Large	digitalocean	$0.080	Unknown	256	1
Ministral 3B (latest)	Mistral	$0.040	$0.040	128K	1
Ministral 3B	azure	$0.040	$0.040	128K	1
Sao10K: Llama 3 8B Lunaris	Meta	$0.040	$0.050	8K	1
GTE Large (v1.5)	digitalocean	$0.090	Unknown	8K	1
Llama-3.2-11B-Vision-Instruct	Meta	$0.049	$0.049	128K	8
Sao10k L3 8B Lunaris	novita-ai	$0.050	$0.050	8K	1
L3 8B Stheno V3.2	novita-ai	$0.050	$0.050	8K	1
MythoMax 13B	kilo	$0.060	$0.060	4K	1
IBM: Granite 4.0 Micro	kilo	$0.017	$0.110	131K	1
Gemma 3 12B	Google	$0.030	$0.100	33K	10
Llama 3.1 8B Instant	Meta	$0.050	$0.080	131K	2
Llama 3 8B	Meta	$0.050	$0.080	8K	1
Gemma 3 27B	Google	$0.027	$0.109	131K	14
Qwen2.5-Coder 32B Instruct	Alibaba (Qwen)	$0.027	$0.109	131K	7
DeepSeek R1 Distill Llama 70B	DeepSeek	$0.027	$0.109	8K	5
baichuan-m2-32b	novita-ai	$0.070	$0.070	131K	1
LiquidAI: LFM2-24B-A2B	kilo	$0.030	$0.120	33K	1
DeepSeek R1 Distill Llama 70B	Meta	$0.030	$0.130	131K	5
GPT OSS 20B	OpenAI	$0.030	$0.140	131K	23
Mythomax L2 13B	novita-ai	$0.090	$0.090	4K	1
Command R7B	Cohere	$0.037	$0.150	128K	2
Command R7B Arabic	Cohere	$0.037	$0.150	128K	1
Arcee AI: Trinity Mini	kilo	$0.045	$0.150	131K	1
Trinity Mini	clarifai	$0.045	$0.150	131K	1
GPT OSS 120B	OpenAI	$0.040	$0.160	131K	33
Qwen3 235B A22B Instruct 2507	Alibaba (Qwen)	$0.100	$0.100	262K	18
Qwen3-235B-A22B-Thinking-2507	Alibaba (Qwen)	$0.100	$0.100	262K	17
Qwen3 30B A3B Instruct 2507	Alibaba (Qwen)	$0.100	$0.100	262K	12
Qwen3 30B A3B Thinking 2507	Alibaba (Qwen)	$0.100	$0.100	262K	7
nvidia-nemotron-nano-9b-v2	NVIDIA	$0.040	$0.160	131K	6
Qwen/Qwen3.5-9B	Alibaba (Qwen)	$0.050	$0.150	262K	6
Phi-4	Microsoft	$0.060	$0.140	128K	4
Reka Edge	kilo	$0.100	$0.100	16K	1
Llama 3.1 8B	Meta	$0.100	$0.100	32K	1
GPT OSS 120B	synthetic	$0.100	$0.100	128K	1
Ministral 8B (latest)	Mistral	$0.100	$0.100	128K	1
Ministral 3B	llmgateway	$0.100	$0.100	131K	1

Top 60 von 414 angezeigt. Im vollständigen Verzeichnis weiter filtern.

Frequently asked questions

How many AI models support offene Gewichte?

414 canonical models in our database currently support offene Gewichte. The list is regenerated on every data refresh, so it always reflects the latest model releases from models.dev.

What is the cheapest model with offene Gewichte?

Whisper Large v3 from scaleway is currently the lowest-priced option, at $0.003 per 1M input tokens and Unknown per 1M output tokens. The full table above is sorted price-ascending.

Which model with offene Gewichte has the largest context window?

Llama 4 Scout 17B Instruct (Meta) leads on context at 3.50M tokens. This may matter if you also need long-document understanding alongside offene Gewichte.

Which models are available on the most providers?

Production-readiness usually correlates with how many independent providers host the same weights. The top three by provider count are: Kimi K2.5 (45), MiniMax-M2.5 (40), GLM-5 (38).

How is offene Gewichte different from a regular LLM?

Open-weight models publish their trained weights publicly. You can self-host on your own GPUs, fine-tune on private data or run offline. Note: open weights ≠ open source — training data and code are usually not released.

How often is this list updated?

Daily. Our data pipeline pulls models.dev once a day, regenerates the canonical model list, and rebuilds these pages so newly released models appear within 24 hours.

Top models with this capability

Whisper Large v3$0.00 in / $0.00 out
Voxtral Small 24B 2507$0.00 in / $0.00 out
KB Whisper$0.00 in / $0.00 out
Multi-QA-mpnet-base-dot-v1$0.01 in / $0.00 out
All-MiniLM-L6-v2$0.01 in / $0.00 out

Other capabilities

Best-of lists you might also want

Pricing comparisons

Vendors in this list

Zuletzt aktualisiert: 2026-05-12

Prices in USD per 1M tokens. Unknown means the provider does not publish per-token pricing.

Data is sourced from models.dev and normalized for comparison. Prices and capabilities may change. Always verify critical production decisions with the provider's official documentation.