swiss-ai/Apertus-v1.5-70B

Hugging Face
TEXT GENERATIONConcurrent Unit Cost:4Model Size:70BQuant:FP8Context Size:32kPublished:Jul 24, 2026License:apache-2.0Architecture:Transformer0.1K Open Weights Gated Warm

Apertus-v1.5-70B by swiss-ai is a 70 billion parameter, decoder-only transformer model with xIELU activation, designed for multilingual, multimodal, and open AI. It supports a massive 262,144 token context length and integrates native image and audio understanding. This model excels in instruction-following, tool-use, and reasoning tasks, featuring a unique 'thinking mode' to enhance performance.

Loading preview...

Apertus-v1.5-70B: Multilingual, Multimodal, and Open AI

Apertus-v1.5-70B is a 70 billion parameter language model from swiss-ai, part of the Apertus 1.5 family, focused on advancing open and transparent AI. This model builds upon Apertus 1.0 with an additional 2 trillion tokens of multimodal pretraining, maintaining the original decoder-only transformer architecture with xIELU activation and AdEMAMix optimizer.

Key Capabilities

  • Multimodal Input: Natively processes images, audio, and text inputs, generating text outputs.
  • Massive Context Length: Supports an extended context window of up to 262,144 tokens.
  • Enhanced Reasoning: Features a unique "thinking mode" that can be enabled to improve performance on complex reasoning tasks.
  • Improved Instruction-Following & Tool Use: Significant advancements in adhering to instructions and integrating external tools/APIs.
  • Fully Open: Provides open weights, open data, open values, and full training details.
  • Multilingual Support: Designed to support a wide variety of languages.

When to Use This Model

  • Multimodal Applications: Ideal for use cases requiring the processing and understanding of interleaved text, image, and audio inputs.
  • Complex Reasoning Tasks: Leverage the "thinking mode" for applications demanding enhanced logical deduction and problem-solving.
  • Long-Context Understanding: Suitable for tasks that benefit from processing extensive amounts of information within a single context.
  • Instruction-Driven Automation: Excellent for scenarios where precise instruction adherence and effective tool integration are critical.
  • Open-Source Development: A strong choice for developers prioritizing fully transparent and open-source AI solutions.

Finetunes
1 models