Model Guide

Model and assistant update in November

The older models Llama 70B, Phi 3 Mini, and Mistral 7B will be discontinued as of November 27, 2025, since more powerful successor models are now available, which we recommend for all current AI applications.

All affected assistants will be automatically switched to Mistral 3.2 on November 27, 2025.

Overview of the available models in the KI-Workplace

Mistral 3.2 is pre-selected as the default in chat and assistant creation.

🚀 Mistral 3.2: Our recommendation

Our most powerful model from European development. The best choice for most scenarios, including the particularly demanding ones.

Whether in-depth analyses, large document collections, precise image processing, or handling complex knowledge: Mistral 3.2 delivers results at the highest level.

Supports text and image processing with contexts of up to 128k tokens (approx. 200 A4 pages).

Note: Response times may be slightly longer during peak periods.

⚡ Gemma 3: Great performance even at peak times

The versatile all-rounder for text and images. Context size up to 32k tokens (approx. 50 A4 pages).

Typical use cases:

• Everyday correspondence
• Analysis of medium-sized documents
• Image understanding
• Tasks with embedded knowledge

Note: Also available during peak periods.

⏱️ Llama 3.1 8B: The fast one, text only

Our fastest text model with 32k tokens (approx. 50 A4 pages). Particularly efficient for simple tasks.

Typical use cases:

• Short texts
• Quick responses
• Routine tasks where speed matters more than precision

Note: Very fast, but not always as accurate as Mistral 3.2 or Gemma 3.

🧩 Legacy-Modelle

For special use cases, primarily for experienced users:

• Llama 3.1 70B – very large language model for highly precise text analyses
• Mixtral 8x7B – mixture-of-experts, strong for technical and domain-specific tasks
• Phi-3 Mini – compact model for simple routine tasks with low resource requirements
• Pixtral 12B – multimodal model for text and image tasks in the mid-performance range