Meta has released its most powerful open-source large language model to date, Llama 3.1 405B, which supports a context length of 128K tokens and was trained on 15 trillion tokens. The total downloads of Llama models have exceeded 300 million, with performance comparable to top models like GPT-4.
This move by Meta is seen as a turning point in the open-source versus closed-source debate. Zuckerberg emphasizes the importance of open-source AI, believing it will become the industry standard. In contrast to OpenAI's secrecy regarding technical details, Meta has published detailed technical papers.
OpenAI subsequently launched a low-cost GPT-4o mini version in response. This marks an escalation in the AI price war in Silicon Valley, with major companies introducing models cheaper than GPT-4.
Smaller models are becoming a tool for AI companies to reduce costs and increase efficiency. They are faster at handling simple tasks, require less storage and energy, and are cheaper. For ordinary users, small models are a cost-effective alternative to large models.
Lower prices stem from reduced costs. By significantly lowering the cost of AI usage, companies like OpenAI hope to expand the range of AI applications. Small models could become a new trend in the AI field.