Apertus 1.5
Apertus 1.5 extends the 8B and 70B Apertus 1.0 models Apertus 1.5 uses the Emu3.5 tokenizer to tokenize images into tokens Apertus 1.5 uses WavTokenizer to tokenize audio into tokens Apertus 1.5 uses FineWeb2-HQ+ as part of its training data Apertus 1.5 uses FinePDFs as part of its training data Apertus 1.5 uses Nemotron-CC as part of its training data Apertus 1.5 adheres to the Apertus Charter Apertus 1.5 uses EvalEval for evaluation reporting Apertus 1.5 uses lmms-eval as part of its evaluation pipeline Apertus 1.5 uses VLMEvalKit as part of its evaluation pipeline Apertus 1.5 uses lm-evaluation-harness as part of its evaluation pipeline Apertus 1.5 uses Hugging Face Community Evals for evaluation reporting Apertus 1.5 maps its evaluation cards to the EU AI Act Apertus 1.5 maps its evaluation cards to the GPAI Code of Practice Apertus 1.5 shows improved performance on the AIME benchmark Apertus 1.5 shows strong performance on Swiss Legal tasks, being on par with leading open‑weight models Apertus 1.5’s multimodal performance is close to Gemma 4 12B EvalEval Coalition collaborates with Apertus 1.5 for transparent benchmarking NVIDIA NeMo AutoModel SUPPORTS_LORA_TRAINING_FOR FLUX 3 Image Klein· Entity summary generated by Graphiti
Relation timeline
BY VALID-FROM DATE- Apertus 1.5COMPETES WITHGemma 4 12B
Apertus 1.5’s multimodal performance is close to Gemma 4 12B
- Apertus 1.5USES EVAL TOOLVLMEvalKit
Apertus 1.5 uses VLMEvalKit as part of its evaluation pipeline
- Apertus 1.5USES EVAL TOOLlm-evaluation-harness
Apertus 1.5 uses lm-evaluation-harness as part of its evaluation pipeline
- EvalEval CoalitionCOLLABORATES WITHApertus 1.5
EvalEval Coalition collaborates with Apertus 1.5 for transparent benchmarking
- Apertus 1.5USES EVAL TOOLEvalEval
Apertus 1.5 uses EvalEval for evaluation reporting
- Apertus 1.5USES EVAL TOOLHugging Face Community Evals
Apertus 1.5 uses Hugging Face Community Evals for evaluation reporting
- Apertus 1.5USES DATA SOURCEFineWeb2-HQ+
Apertus 1.5 uses FineWeb2-HQ+ as part of its training data
- Apertus 1.5EXTENDSApertus 1.0
Apertus 1.5 extends the 8B and 70B Apertus 1.0 models
- Apertus 1.5ADHERES TOApertus Charter
Apertus 1.5 adheres to the Apertus Charter
- Apertus 1.5USES TOKENIZEREmu3.5 tokenizer
Apertus 1.5 uses the Emu3.5 tokenizer to tokenize images into tokens
- Apertus 1.5PERFORMS BETTER INAIME
Apertus 1.5 shows improved performance on the AIME benchmark
- Apertus 1.5USES DATA SOURCENemotron-CC
Apertus 1.5 uses Nemotron-CC as part of its training data
- Apertus 1.5MAPS TOEU AI Act
Apertus 1.5 maps its evaluation cards to the EU AI Act
- Apertus 1.5MAPS TOGPAI Code of Practice
Apertus 1.5 maps its evaluation cards to the GPAI Code of Practice
- Apertus 1.5PERFORMS ONSwiss Legal
Apertus 1.5 shows strong performance on Swiss Legal tasks, being on par with leading open‑weight models
- Apertus 1.5USES TOKENIZERWavTokenizer
Apertus 1.5 uses WavTokenizer to tokenize audio into tokens
- Apertus 1.5USES EVAL TOOLlmms-eval
Apertus 1.5 uses lmms-eval as part of its evaluation pipeline
- Apertus 1.5USES DATA SOURCEFinePDFs
Apertus 1.5 uses FinePDFs as part of its training data
- Apertus 1.5SUPPORTS LORA TRAINING FORFLUX.3 Image Klein
NVIDIA NeMo AutoModel SUPPORTS_LORA_TRAINING_FOR FLUX 3 Image Klein
Relations are facts Graphiti extracted from reports, not verbatim quotes. An end date means later reporting showed the relation no longer holds.