Main use cases: Can be used for any task that requires speech generation. For example summaries, chatbots, voicebots, but also intent or sentiment recognition.  

 
Input length: Two different models with 4,096 tokens (approx. 3,072 words) or 16,385 tokens (approx. 12,288 words)  

 
Languages: 95 natural languages  

 
Model size: 110 billion parameters  

 
Evaluated in: VIER summarization, VIER intent detection, VIER dialog

Main use cases: Can be used for any task that requires speech generation. Very high quality for summaries, chat applications, concern and sentiment recognition as well as the generation of creative content, coding or general knowledge. Is able to evaluate the performance of other models in dialog tasks. Can also process images as input.  

 
Input length: Three different models with 8,192 tokens (approx. 6,144 words), 32,769 tokens (approx. 24,576 words) and 128,000 tokens (over 300 pages of continuous text)  

 
Languages: 95 natural languages, is better than GPT-3.5 Turbo in at least 26 languages  

 
Model size: ~1.8 trillion parameters  

 
Evaluated in: VIER summarization, VIER dialogs

Main use cases: The main task of the model is language generation. It can also recognize concerns and sentiment and formulate texts more optimally (e.g. simplified) if it is given a few examples. For chat applications, it needs to be fine-tuned beforehand so that there are no unwanted responses with discriminatory content or biases.  

 
Input length: 2048 tokens (approx. 1536 words)  

 
Languages: English, German, French, Italian and Spanish  

 
Model size: ~70 billion parameters  

 
Evaluated in: VIER summarization

Main use cases: Can be used similarly to GPT-4 for any form of language generation, for example for creative content creation, text summarization, text editing, in-depth dialog, understanding complex contexts or coding.  

 
Input length: 100,000 tokens (approx. 300 pages of continuous text)  

 
Languages: best in English, but also possible in at least 43 other languages  

 
Model size: ~130 billion parameters  

 
Evaluated in: VIER summarization

Main use cases: A general language generation model that can be used for any form of language generation, such as creative content creation, text summarization, text editing, advanced dialog or understanding complex contexts.  

 
Input length: 9,000 tokens (approx. 6750 words)  

 
Languages: Mainly English. Also Spanish and French. To a lesser extent also German, Italian, Portuguese and possibly other languages.  

 
Model size: ~93 billion parameters (estimate, as unpublished)  

 
Evaluated in: VIER summarization

Main use cases: Can generate speech in various contexts, for example for summaries, coding, dialog guidance, text editing or translation. Can also be used for generating creative content.  

 
Input length: 8196 tokens (approx. 6100 words)  

 
languages: more than 100 natural languages  

 
Model size: ~340 billion parameters  

 
Evaluated in: -

Main use cases: Model for speech generation that can be used for general chat tasks. It has been specially trained for business use cases such as summarizing or (re)formulating texts as well as information extraction or intent recognition.  

 
Input length: 4096 tokens (approx. 3072 words)  

 
Languages: Mainly English. To a lesser extent also German, French, Italian, Spanish and Arabic.  

 
Model size: ~52 billion parameters  

 
Evaluated in: -

Main use cases: An embedding model that compares the input text with text references and calculates the similarity. This can be used, for example, to implement search functions (in knowledge databases), intent recognition and text classification.  

 
Input length: 8191 tokens (approx. 6143 words)  

 
Languages: Mainly English. Also Spanish, French, German, Italian, Portuguese, Mandarin and probably many others.  

 
Model size: ~350 million parameters  

 
Evaluated in: VIER intent detection

Main use cases: An embedding model that compares the input text with text references and calculates the similarity. This can be used, for example, to implement search functions (in knowledge databases), intent recognition and text classification.  

 
Input length: 2048 tokens (approx. 1536 words)  

 
Languages: English, German, French, Italian and Spanish  

 
Model size: ~13 billion parameters  

 
Evaluated in: VIER intent detection

Main use cases: Model for language generation that can be used for general chat tasks. The model can also recognize the customer's most important intents and summarize texts in a simple way. However, the quality in these tasks is correspondingly poorer than with the larger alternatives.  

 
Input length: 4096 tokens (approx. 3072 words)  

 
Languages: English  

 
Model size: ~7 billion parameters  

 
Evaluated in: VIER intent detection

Main use cases: Model for speech generation, which can be used for general chat tasks and communicates in a human-like manner. The model is based on Llama2 and has been improved with chat conversations. Even if it can basically recognize the most important intents - and summarize texts, for example - the quality is limited by the size.  

 
Input length: 4000 tokens (approx. 3000 words)  

 
languages: predominantly English  

 
Model size: ~7 billion parameters  

 
Evaluated in: VIER summarization

Main use cases: Model for speech generation, which can be used for translations, text summaries, sentiment analysis or intent recognition. The quality of language generation lags behind larger, more modern models, while intent recognition, for example, is similarly good.  

 
Input length: 512 tokens (approx. 384 words) is basic, trained up to 2048 tokens (approx. 1536 words)  

 
Languages: English, French, Romanian, German  

 
Model size: ~11 billion parameters  

 
Evaluated in: VIER intent detection

Main use cases: Model for speech generation, which can be used for translations, text summaries, sentiment analysis or intent recognition. The quality of language generation lags behind larger, more modern models, while intent recognition, for example, is similarly good.  

 
Input length: 512 tokens (approx. 384 words) is basic, up to 2048 tokens (approx. 1536 words) trained  

 
Languages: English, French, Romanian, German  

 
Model size: ~3 billion parameters  

 
Evaluated in: VIER intent detection

Main use cases: Model that has been fine-tuned for intent recognition and text classification. It is based on the RoBERTa variation of the basic BERT model and can recognize intents even in complex emails if only the name of the intent is specified (zero shot).  

 
Input length: 512 tokens (approx. 384 words)  

 
Languages: English, French, German, Spanish, Greek & 10 others  

 
Model size: ~355 million parameters  

 
Evaluated in: VIER intent detection

Main use cases: A model for language generation that has been fine-tuned for the general answering of questions (instruction-tuned). This means that the model can be used for various tasks such as intent recognition, summarization, (re-)formulation or text classification.  

 
Input length: 8192 tokens (approx. 6144 words)  

 
Languages: predominantly English  

 
Model size: ~30 million parameters  

 
Evaluated in: -

Main use cases: Model for language generation that has been fine-tuned for the general answering of questions (instruction-tuned). This means that the model can be used for various tasks such as concern recognition, summarization, (re-)formulation or text classification. This model is based on the open source LLM Pythia from EleutherAI. The basic model is only trained for English, so it cannot currently be used for German-language use cases.  

 
Input length: 2048 tokens (approx. 1536 words)  

 
Languages: English  

 
Model size: ~12 million parameters  

 
Evaluated in: -

Main use cases: A multilingual model based on a further development of the BERT architecture. It was explicitly trained for concern recognition and text classification without examples (zero shot).  

 
Input length: 512 tokens (approx. 384 words) - theoretically 24,528 tokens, but considerable speed losses are to be expected above 512 tokens.  

 
Languages: Evaluated for 15 languages, including English and German. To a lesser extent 85 other languages.  

 
Model size: ~86 million parameters  

 
Evaluated in: VIER intent detection

Main use cases: A general language model based on the Transformer architecture. It was explicitly trained for intent recognition and text classification without examples (zero shot).  

 
Input length: 1024 tokens (approx. 768 words)  

 
Languages: predominantly English  

 
Model size: ~407 million parameters  

 
Evaluated in: VIER intent detection

Main use cases: A general language generation model based on Llama2 and fine-tuned for use in chat conversations. It can be used for dialog systems, text summarization or concern detection in English.  

 
Input length: 4096 tokens (approx. 3072 words)  

 
languages: predominantly English  

 
Model size: ~13 billion parameters  

 
Evaluated in: -

Main use cases: Model for language generation that has been fine-tuned to answer general questions (instruction-tuned). This means that the model can be used for various tasks such as concern recognition, summarization, (re-)formulation or text classification. However, possible use cases are limited by the rather short input length.  

 
Input length: 2048 tokens (approx. 1536 words)  

 
Languages: Mainly English. Also German, Spanish and French. To a lesser extent also Italian, Portuguese, Polish, Dutch, Romanian, Czech and Swedish.  

 
Model size: ~40 billion parameters  

 
Evaluated in: -

Main use cases: Model for speech generation that can be used for general chat tasks. The model can be used for dialog systems, concern recognition or summaries. However, possible use cases are limited by the rather short input length.  

 
Input length: 2048 tokens (approx. 1536 words)  

 
Languages: Mainly English. Also German, Spanish, French. To a lesser extent also Italian, Portuguese, Polish, Dutch, Romanian, Czech and Swedish  

 
Model size: ~180 billion parameters  

 
Evaluated in: -

Main use cases: Model for language generation that can be used for general chat tasks. The model can also recognize the customer's most important concerns and summarize texts in a simple way. The quality is better than the smaller alternative (Llama2-7B-Chat), depending on the size.  

 
Input length: 4096 tokens (approx. 3072 words)  

 
Languages: Mainly English. To a lesser extent also Spanish, French, German, Italian, Portuguese, Dutch, Russian, Chinese (Mandarin), Japanese, Korean, Arabic, Hebrew, Hindi, etc.  

 
Model size: ~13 billion parameters  

 
Evaluated in: -

Main use cases: Guanaco 65B is interesting because, thanks to a very efficient approach (QLoRA), a model of this size can be fine-tuned on very limited hardware. The result is a model for speech generation that solves general chat tasks in good quality. However, due to the license of the basic model (Llama1-65B), commercial use is excluded.  

 
Input length: 2048 tokens (approx. 1536 words)  

 
Languages: Mainly English. Additionally Spanish. To a lesser extent also Russian, German, French, Chinese, Thai, Brazilian Portuguese, Catalan, etc.  

 
Model size: ~65 billion parameters  

 
Evaluated in: -  

 
Only non-commercial use permitted!