EmbeddingGemma 2: an open, lightweight multimodal embedding model
deepmind.google·6d ago
AI Engineering
Hacker News·Sep 8·5 min read
Mercury 2.5
TL;DR
Mercury 2.5 has been released as the most advanced diffusion language model, boasting a 40% increase in intelligence over its predecessor, Mercury 2. It is designed for low-latency, cost-effective applications across various industries.
✦ Why It Matters
Consider integrating Mercury 2.5 into your applications for improved performance in latency-sensitive tasks.
Key Takeaways
- Test Mercury 2.5 for search and voice applications to enhance user experience.
- Leverage the introductory pricing to evaluate its capabilities without high costs.
- Utilize the Inception API for easy integration into existing systems.
- Explore Mercury Voice for optimized performance in voice agent applications.
Full Summary
Mercury 2.5 is the latest version of a diffusion language model, designed to enhance performance in latency-sensitive applications like search, voice, and coding. It boasts a 40% increase in intelligence compared to its predecessor, Mercury 2, and operates at speeds of 1,107 tokens per second on NVIDIA GPUs.
The model supports a context of 260,000 tokens and is priced competitively, with an introductory discount. Feedback from real-world usage has informed its development, leading to improvements in reasoning, tool calls, and JSON schema alignment.
Early adopters have reported significant reductions in response times and costs, making it a strong contender against other leading models like GPT-5.6 Luna and Claude Haiku 4.5.