Mercury 2

Inception Labs has developed Mercury 2, a new language model designed to be the world's fastest at reasoning. Its main goal is to make artificial intelligence feel instant when used in real applications.

Benefits

Mercury 2 works differently from other language models. Instead of generating text one word at a time, it creates responses by refining them in steps, producing multiple words at once. This makes it over five times faster than traditional models. It offers reasoning abilities similar to slower models but with much quicker response times, which is great for applications that need to be fast.

Use Cases

This model is especially helpful for applications where speed is very important and affects how users experience the product. This includes:

• Coding and editing:It can help with things like code suggestions and interactive coding tools, providing quick feedback that keeps programmers working smoothly.
• Agentic loops:For tasks that require many AI steps, Mercury 2's speed per step makes the whole process more efficient and leads to better results.
• Real-time voice and interaction:It allows for high-quality responses within the short time limits needed for natural speech, improving experiences with AI avatars and voice assistants.
• Search and RAG pipelines:It can add reasoning to search processes without slowing them down, helping to find, rank, and summarize information more effectively.

Pricing

Mercury 2 costs $0.25 for every 1 million input tokens and $0.75 for every 1 million output tokens.

Vibes

Users have noted its speed and effectiveness. For example, Adrian Witas from Viant mentioned its use in improving campaign execution, and Suchintan Singh from Skyvern stated it is at least twice as fast as GPT-5.2. Timo Selvaraj from SearchBlox found it practical for real-time AI in their search product. Max Sapo from Happyverse AI and Oliver Silverstein from OpenCall highlighted its role in creating natural voice interactions.

Additional Information

Mercury 2 is now available and works with the OpenAI API, making it easy to add to existing systems. Inception Labs also offers partnerships for businesses to test the model and get help with evaluating its performance for their specific needs.