.png)
December 8, 2024
Meta has released a new version of its large language model, Llama 3.3. The most notable feature of this iteration is its improved efficiency: Llama 3.3 70B achieves performance comparable to the significantly larger Llama 3.1 405B, but at considerably lower cost. This is made possible by advancements in the alignment process and online RL techniques. In direct comparison with other models such as Google's Gemini 1.5 Pro, OpenAI's GPT-4o, and Amazon's Nova Pro, Llama 3.3 70B performs competitively in various benchmarks, including MMLU, MATH, and GPQA Diamond.
The improved cost efficiency of Llama 3.3 70B is a crucial factor for developers and businesses. With a price of approximately $0.10 USD per million input tokens and $0.40 USD per million output tokens, the model is significantly more affordable than the larger 405B model. This cost reduction, while maintaining performance, makes Llama 3.3 accessible to a wider audience and allows smaller companies and developers to utilize powerful AI models.
Llama 3.3 70B is a text-based model with a context length of up to 128,000 tokens. It supports multiple languages, including English, German, French, Italian, Portuguese, Hindi, Spanish, and Thai. The knowledge cutoff is December 2023. The model is available on platforms like Hugging Face, where it can also be tested for free. Meta emphasizes the strengths of Llama 3.3 70B in multilingual applications and tasks with long context.
Meta's focus on open-source AI models with Llama 3.3 underscores the growing trend towards more transparent and accessible AI solutions. The combination of high performance and cost efficiency could change the competitive landscape in the field of large language models and increase pressure on other providers to offer comparable performance at lower prices. The development of Llama 3.3 is another step towards the democratization of AI technologies and could lead to a wider application of AI across various industries.
Developments in the field of large language models, such as the release of Llama 3.3, highlight the importance of platforms like Mindverse, which support companies in integrating and applying AI technologies. Mindverse offers a comprehensive suite of AI tools for text, images, research, and more. In addition, Mindverse develops customized solutions such as chatbots, voicebots, AI search engines, and knowledge systems, which help companies fully realize the potential of AI. The continuous improvement in the performance and efficiency of AI models opens up new possibilities for innovation and the development of intelligent applications in the future.
Bibliographie: - https://buttondown.com/ainews/archive/ainews-meta-llama-33-405bnova-pro-performance-at/ - https://artificialanalysis.ai/models/llama-3-3-instruct-70b - https://artificialanalysis.ai/models/llama-3-3-instruct-70b/providers - https://www.mediapost.com/publications/article/401697/meta-launches-new-cost-efficient-llama-ai-model.html?edition=136654 - https://www.llama.com/ - https://beebom.com/meta-llama-3-3-70b-model-matches-405b-performance/ - https://www.reddit.com/r/LocalLLaMA/comments/1h85tt4/meta_releases_llama33_70b/ - https://www.benzinga.com/24/12/42376972/metas-llama-33-70b-the-new-ai-model-that-outperforms-google-openai-and-amazon-with-unmatched-efficiency - https://groq.com/new-ai-inference-speed-benchmark-for-llama-3-3-70b-powered-by-groq/ - https://www.datacamp.com/blog/llama-3-3-70b ```