In December 2024, the landscape of large language models (LLMs) saw significant developments, with major releases from Amazon, Meta, and Google. Amazon introduced the Amazon Nova family of multi-modal models, competing with Google's Gemini 1.5 series. Meta released Llama 3.3 70B-Instruct, a model comparable to their larger 405B model, which can run on laptops and is available from API providers like Groq and Cerebras. The efficiency of these models has improved dramatically, leading to reduced costs for proprietary hosted LLMs. Google made several impactful updates, including the experimental gemini-exp-1206 and the official Gemini 2.0 Flash, which featured impressive streaming capabilities. They also introduced Gemini 2.0 Flash “Thinking mode,” showcasing inference scaling techniques.
OpenAI's 12 Days of OpenAI event featured notable releases, including the full o1 model, o1-pro, and various enhancements to ChatGPT, such as Advanced Voice with video streaming and a new WebRTC streaming API. The competition between Google and OpenAI has intensified, with Google making strides in the LLM space. The blog entries and releases highlight the ongoing advancements and tools being developed to leverage these powerful models, including plugins for accessing different LLMs and tools for data enrichment and querying.
Key takeaways:
```html
Amazon and Google have released new multi-modal LLMs, with Amazon introducing the Nova family and Google launching Gemini 2.0 Flash.
Meta's Llama 3.3 70B-Instruct model is highly efficient, allowing users to run a GPT-4 class model on a laptop.
The competition between LLM vendors has driven down the cost of proprietary hosted LLMs significantly.
OpenAI's 12 Days of OpenAI event featured several notable releases, including the o1 model and a new WebRTC streaming API.
