Bitget
Bitget

# AI

Meta’s Llama 3.1 405B Model Outperforms GPT-4 in Leaked Benchmarks

Meta, Llama

KEYTAKEAWAYS

  • Leaked benchmarks reveal Llama 3.1 405B outperforms GPT-4 in multiple tests.
  • This marks the first instance of an open-source model surpassing a leading closed-source LLM.

CONTENT

Early benchmark data for Meta AI’s upcoming Llama 3.1 models, including the massive 405B parameter version, shows surprising performance gains. The open-source model potentially surpasses OpenAI’s GPT-4 in several key AI benchmarks.


 

 

In a significant development for the artificial intelligence community, leaked benchmarks for Meta AI’s upcoming Llama 3.1 405B model suggest it may outperform OpenAI’s GPT-4, the current industry leader. This revelation, posted on the LocalLLaMA subreddit, marks a potential milestone as the first instance of an open-source model surpassing a leading closed-source large language model (LLM).

 

 

 

 

 

Meta introduced Llama 3 in April 2024 as a new generation of open-source LLMs, with initial releases of 8B and 70B parameter models. The company announced plans for a more ambitious 400 billion parameter model, which is still in training. The leaked data includes benchmarks for the 8B, 70B, and the massive 405B parameter models.

 

Meta, LLM

 

The benchmarks show Llama 3.1 outperforming GPT-4 in several crucial areas, including GSM8K, Hellaswag, BoolQ, and various MMLU categories (humanities, STEM, and other). However, it lags in HumanEval and MMLU-social sciences tests, indicating room for improvement.

 

Meta, LLM

 

It’s important to note that these results reflect the base models’ performance. The upcoming instruction-tuned versions are expected to yield even better results across various benchmarks.

 

This development underscores the growing influence and capability of open-source AI initiatives. Meta’s commitment to the open-source ethos, as stated in their blog post launching Llama 3, aims to democratize access to advanced AI models and tap into the global developer community’s collective intelligence.

 

While GPT-5 may challenge Llama 3.1’s emerging dominance in the future, the impressive performance of this open-source model against GPT-4 highlights the potential for widespread innovation and enhancement in AI technology.

 

>> Also read: What is GPT-4o? OpenAI’s Most Advanced AI Model Yet

 

 

▶ Buy Crypto at BingX

 

Sign up to claim 5,000+ USDT in rewards & 20% off trading fees!

 

BingX


Looking for the latest scoop and cool insights from CoinRank? Hit up our Twitter and stay in the loop with all our fresh stories!


DISCLAIMER

CoinRank is not a certified investment, legal, or tax advisor, nor is it a broker or dealer. All content, including opinions and analyses, is based on independent research and experiences of our team, intended for educational purposes only. It should not be considered as solicitation or recommendation for any investment decisions. We encourage you to conduct your own research prior to investing.

 

We strive for accuracy in our content, but occasional errors may occur. Importantly, our information should not be seen as licensed financial advice or a substitute for consultation with certified professionals. CoinRank does not endorse specific financial products or strategies.


WRITER’S INTRO

CoinRank Exclusive brings together primary sources from various fields to provide readers with the most timely and in-depth analysis and coverage. Whether it’s blockchain, cryptocurrency, finance, or technology industries, readers can access the most exclusive and comprehensive knowledge.


NEWSLETTER

SUBSCRIBE

CoinRank