The most popular LLMs and their applications - 7 GenAI models you should know

149 GenAI models are currently on the market. According to the Standford AI Index Report (2024) there are over 109 companies in the USA alone that provide Large Language Model (LLM) services.

Staying up to date in this space is challenging, especially with over 12 new model releases in just the past week. In our previous newsletter, we delved into the characteristics of LLMs. Today, we explore some of the most popular models and what sets them apart.

The LLM Ecosystem - The 7 LLM Models I follow in 2024

It would absorb way too much time to elaborate all the 149 LLMs. To provide clarity, I've chosen to spotlight the 7 LLM models that, in my view, are the most prominent in the current landscape. While this selection may evolve over time due to the dynamic nature of the market, these models represent significant players as of now.

I start with the most prominent ones: OpenAI's GPT Series, Google's Gemini, Meta's Llama, Anthropic's Claude, Cohere's Command, Mistral's Mixtral, and Aleph Alpha's Luminous. I focus on the strengths and weaknesses of each provider to give you an overview and first level of comparison to help you make a more informed decision of what model is the best fit for you:

GPT-Series OpenAI : Most popular and one of the best-performing LLM models. Known for versatility and performance in text generation, language understanding, and various NLP tasks. The GPT series offers a range of models like GPT-3.5 Turbo and GPT-4.

• Strengths: Multimodality and performance. These models excel in text generation, language understanding, and various NLP tasks. The most popular and widely used model for chatbots, content creation, and natural language processing.
• Weakness: Cost, latency and transparency could be challenged by competing models.

Gemini Google: OpenAI competitor. Built for multimodality. Strong reasoning across text, images, audio, video, and code. Gemini comes in different versions (Nano, Pro, Ultra).

• Strengths: Fierce competitor of OpenAI. Balanced model size with practical usability, built for multimodality. Bring a lot of firepower in terms of unique data and capital to spend.
• Weakness: Struggling a bit to catch up to OpenAI's head start and model performance. (Google's Bad Gemini Rollout)

Llama Meta : One of the most popular open-source models. Comes in model sizes 8B and 70B. Llama 3 just got released on April 18th, 2024.

• Strengths: The latest release claims similar and better performance levels as GPT-4. One of the most used Open-source models. Ideal for research, experimentation, and community collaboration.
• Weakness: The "Open" in open-source is being disputed right now.

Claude Anthropic: A versatile model suitable for a wide range of tasks, from lightweight actions to complex analysis. Claude 3 claims on par or better performance than GPT-4.

• Strengths: Versatile and suitable competitor to GPT-4 and Gemini Ultra. Used for a wide range of tasks.
• Weakness: Competing with OpenAI and Google will require a lot of capital. Anthropic raised USD 7.3 billion over the last year, let's see if it is enough.

Command R, etc. Cohere : Practical focus for business applications, offering efficiency and accuracy in real-world scenarios. Three different main models (Command, Rerank, Embed).

• Strengths: Cost-effective smaller models especially trained for business applications like Retrieval Augmentation Generation (RAG), Embedding, and Reranking.
• Weakness: From our experience, the market for customized business applications is still early but with huge potential. (There are not many Klarna examples, yet)

Mixtral Mistral AI : The French LLM solution. Offering efficient and cost-effective models in three sizes (Mixtral 7B, 8x7B, 8x22B).

• Strengths: Balances efficiency without sacrificing performance. Especially the smaller models offer very good performance compared to size. One of the popular European LLM providers.
• Weakness: With the latest development, they are shifting away from the openness they once embraced. Microsoft just invested heavily in them, which could be conflicting with their "European" branding.

Luminous Aleph Alpha : The German LLM solution. Special focus on the industry's use of LLMs. Available in three different versions (base, extended, supreme).

• Strengths: Handles diverse tasks while maintaining traceability and explainability, multilingual and multimodal. European solution for more transparency and with a special focus on industrial applications.
• Weakness: Capital will be crucial to keep up in this compute and cost-intensive industry. Aleph Alpha closed a EUR 500 Million round in 2023 and is probably soon in need of another round.

Three things to remember

• Selecting the right LLM is a critical step in leveraging artificial intelligence for language processing tasks effectively.
• Carefully considering factors such as task requirements, model size, resource constraints, and ethical considerations, can help you to make an informed decision that aligns with your objectives.
• Remember, the journey towards selecting the perfect LLM is as enriching as the destination itself.

My Advice

Start with one of the most popular models and take it as a baseline for the performance that is possible without much individualisation. At Connect AI , we usually begin with a GPT-4 Turbo Model from OpenAI and then tailor our approach based on our clients' specific needs and preferences.

Happy modelling!