OpenAI, Anthropic, Google, and Meta are now releasing major model updates every few months. The pace of improvement is compressing what used to take years into quarters.

In 2023, a major LLM release was a big event. In 2025, they're almost routine.

OpenAI released o1, o3, and GPT-4o variants in rapid succession. Anthropic's Claude 3.5 Sonnet was widely regarded as the best coding model. Google's Gemini 2.0 Flash is now competitive with the heavyweights at a fraction of the cost.

**What's changed:** Open-source models (Llama, Qwen, Mistral) have closed the gap dramatically. Many researchers argue open-source is now within 6 months of frontier.

**The business question:** How do you build a moat when the underlying model is a commodity?

The Big Labs Are Releasing Models Faster Than Ever