Recently, the AI industry has witnessed a peak competition between speed and performance. The well-known innovative company Inception Labs has officially launched its new generation diffusion large model, Mercury3. As an upgraded version of the original model, the number of parameters in this model has increased by more than 40 times, and its overall capabilities have achieved a qualitative leap.
In several benchmark tests that have attracted significant attention in the industry, Mercury3 successfully surpassed GPT-6Luna and Claude Haiku5.5. Not only did its scores exceed those of the competitors, but it also maintains an amazing rate of thousands of tokens per second in terms of generation speed. The highest operating speed is even approximately 8 times faster than that of GPT-6Luna.
In terms of technical architecture and parameter configuration, the context window of Mercury3 reaches 1 million Tokens, allowing it to easily accommodate massive amounts of information. What is even more remarkable is that its pricing is much lower than that of mainstream large models currently on the market. Through a unique parallel generation technology approach, this model has fully demonstrated that the diffusion approach still retains an overwhelming speed advantage after scaling up, and it is perfectly capable of handling various tasks such as knowledge问答, complex instruction parsing, and software engineering.
Currently, Mercury3 has been implemented in various cutting-edge scenarios such as browser agents and real-time search via smart glasses. In agent systems that require multiple steps and frequent calls, it serves as a cost-effective fast execution model that efficiently handles high-frequency core tasks like retrieval and information extraction. This significantly improves overall task completion efficiency while reducing development and usage costs.