Hosted Models

Choose from our curated selection of open-source LLMs, all running on sovereign GPU infrastructure in Germany.

tngtech/DeepSeek-TNG-R1T2-Chimera
Powerful reasoning, for complex coding. Model merge combining weights from DeepSeek-R1, DeepSeek-R1-0528 and DeepSeek-V3-0324.
chatcodefimtoolsreasoningmultilingualenglishgerman
Production
160K max tokens
160K input
52 tok/s
zai-org/GLM-5.2Default
Frontier open-weights model rivaling Claude Opus 4.8 and GPT-5.5. #1 open model on the Artificial Analysis Intelligence Index, 1M-token context, built for long-horizon agentic coding.
chatcodetoolsreasoningmultilingualenglishgerman
FlagshipProduction
1105K max tokens
1105K input
28 tok/s
nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16
NVIDIA's flagship reasoning model. 550B sparse MoE (55B active), strong agentic and tool-use capabilities, optimized for enterprise inference.
chatcodetoolsreasoningmultilingualenglishgerman
Production
128K max tokens
128K input
38 tok/s
deepseek-ai/DeepSeek-V4-Flash
Experimental flash variant of DeepSeek-V4. For testing and evaluation purposes.
chatcodetoolsreasoningmultilingualenglishgerman
Experimental
391K max tokens
391K input
18 tok/s
google/gemma-4-31B-it
Lightweight multimodal agent. Well-suited for reasoning, agentic workflows, coding, and multimodal understanding.
chatcodetoolsimagesreasoningmultilingualenglishgerman
Experimental
119K max tokens
119K input
23 tok/s
zai-org/GLM-5.1
Powerful for agentic coding. Significantly stronger coding capabilities than GLM-5, built for long-horizon agentic tasks.
chatcodetoolsreasoningmultilingualenglishgerman
Experimental
198K max tokens
198K input
32 tok/s
Qwen/Qwen3.5-397B-A17B-FP8
Multimodal reasoning and visual agent. Unified vision-language foundation with global linguistic coverage.
chatcodetoolsimagesreasoningmultilingualenglishgerman
ProductionLong-term Stable
128K max tokens
128K input
74 tok/s
openai/gpt-oss-120b
General purpose, high reasoning. Open-weight model for powerful reasoning, agentic tasks, and versatile developer use.
chatcodetoolsreasoningmultilingualenglishgerman
ProductionLong-term Stable
128K max tokens
128K input
108 tok/s
zai-org/GLM-4.7-Flash
Light-weight coding and agentic browsing. Strongest 30B-class MoE model balancing performance and efficiency.
chatcodetoolsreasoningmultilingualenglishgerman
Production
198K max tokens
198K input
98 tok/s
mistralai/Mistral-Small-3.2-24B-Instruct-2506
Fast, for agents and vision-understanding. Low-latency function calling, programming, and visual understanding.
chatcodetoolsimagesmultilingualenglishgerman
Production
29K max tokens
29K input
56 tok/s
Qwen/Qwen3.6-35B-A3B-FP8
Lightweight multimodal agent. Unified vision-language foundation with global linguistic coverage.
chatcodetoolsimagesreasoningmultilingualenglishgerman
Experimental
256K max tokens
256K input
173 tok/s
tngtech/tng-codepilot-v2
Default code-completion model. Based on Qwen2.5-Coder-32B, finetuned for relevant completions with additional file context.
codefimmultilingualenglishgerman
TNG Fine-tunedProduction
28K max tokens
28K input
44 tok/s
tngtech/tng-codepilot
Mature code-completion model. Based on StarCoder2-15b, supporting 600+ programming languages.
codefimmultilingualenglishgerman
TNG Fine-tunedProduction
8K max tokens
8K input
48 tok/s
allenai/olmOCR-2-7B-1025-FP8
Best for optical character recognition. Fine-tuned from Qwen2.5-VL-7B-Instruct using GRPO RL training.
codetoolsimagesmultilingualenglishgerman
63K max tokens
62K input
142 tok/s
tngtech/olmocr-7B-faithful
Best for optical character recognition. TNG-trained version of olmOCR using the olmOCR-mix-0225 dataset.
codetoolsimagesmultilingualenglishgerman
8K max tokens
8K input
62 tok/s
Qwen/Qwen3Guard-Gen-8B
Safety moderation model for content guardrails.
multilingualenglishgerman
32K max tokens
32K input
84 tok/s