?? What a month for new LLMs!

Hi everyone ??

I'm Simona Mazzarino, data scientist at Clearbox AI, and I'm happy to meet you here again, especially because there is big news from the AI world.

Let's welcome GTP-4o and Llama 3

In less than one month, two state-of-the-art LLMs were released: OpenAI’s GPT-4o and Meta’s Llama 3, so today I'd like to share with you a brief comparison between them.

GPT-4o offers real-time, multimodal interaction, excels in multilingual tasks, and provides enhanced efficiency and safety features.

Meta’s Llama 3, with models scaling up to 70 billion parameters, emphasizes open-source innovation and scalability, pushing the boundaries of what open-source AI can achieve.

A comparison

Comparing the two, GPT-4o supports multimodal input and output, including text, images, audio, and video snapshots, though these features are not yet available via API.

Llama 3, on the other hand, currently lacks multimodal capabilities but Meta plans to add them later this year. Regarding context length, GPT-4o takes 128K token window, while Llama 3 8K- is efficiently utilized.?

In few words

On benchmarks, GPT-4o impresses with its speed despite added capabilities, while Llama 3 performs well considering its size (8B). Pricing shows GPT-4o as expensive, whereas Llama 3 offers the best performance/cost ratio for most tasks.?

In conclusion, both GPT-4o and Llama 3 represent significant advancements in the field of language models, each with unique strengths. Choosing the right model will depend on specific needs, balancing advanced features, context handling, and cost efficiency.

Have you already tried these models? What was your experience?

Source: Medium, Meta, Open AI

We would love your feedback about our newsletter and thoughts on the articles we share! Please feel free to contact me anytime about it or write to [email protected].

See you soon!

Simona