Benchmark results published by reputable AI performance benchmarking organizations

Meta Unveils Llama 3.3: 70B Parameter Model Outperforms Competitors

Dec 6, 2024, 05:31 PM

Meta Platforms Inc. has released Llama 3.3, a new open-source language model with 70 billion parameters, which matches the performance of its larger predecessor, Llama 3.1, with 405 billion parameters. This model offers significant improvements in efficiency and cost-effectiveness, making it suitable for text-based applications like synthetic data generation. Llama 3.3 supports eight languages, has a 128,000 token context window, and was trained on over 15 trillion tokens. It outperforms several competitors including Google's Gemini 1.5 Pro, OpenAI's GPT-4o, and Amazon's Nova Pro in various benchmarks, with scores of 50.5% on GPQA Diamond (CoT) and 77.0% on Math (CoT). The model is now available for developers on platforms like Meta, Hugging Face, and Groq, with partnerships extending its reach to services like Fireworks AI and Databricks.