Inception debuts Mercury 3, a diffusion LLM that it says tops GPT-6 Luna, Claude Haiku 5.5, and GLM-5.3 Flash on agentic benchmarks while being 4x to 8x faster

AI models that generate responses up to 8x faster are now available at potentially lower costs.

Inception has launched Mercury 3, a diffusion-based large language model (LLM) that it claims outperforms models like GPT-6 Luna and Claude Haiku 5.5 on agentic benchmarks. The company states Mercury 3 achieves speeds of 1,000 tokens per second, making it significantly faster than traditional autoregressive LLMs. This speed is attributed to its ability to parallelize inference, generating and refining multiple tokens simultaneously, which leverages GPU capabilities more efficiently. Mercury 3 is priced competitively, with output tokens costing $0.35 per million, representing a 35% discount compared to some leading models, though this is an introductory offer. Inception CEO Stefano Ermon highlighted that the model's primary advantage is speed, not just cost, positioning it as a flagship product. The development of diffusion LLMs for text generation marks a notable shift, with Google DeepMind also exploring similar technology with its DiffusionGemma model.

The introduction of Inception's Mercury 3 challenges the prevailing industry strategy of scaling larger traditional LLMs, suggesting that alternative architectures like diffusion models can achieve comparable or superior performance with greater speed and efficiency. This could lead to a significant recalibration of AI development priorities, shifting focus from sheer model size to architectural innovation and optimized inference techniques. Companies that continue to invest heavily in massive autoregressive models might face increasing pressure if diffusion LLMs prove to be a more sustainable and cost-effective path forward for high-performance AI.

This development may signal a broader trend toward more efficient AI architectures that better utilize existing hardware, potentially reducing the massive compute requirements and associated costs that have characterized the current AI boom. If diffusion LLMs become mainstream, it could democratize access to advanced AI capabilities, enabling a wider range of applications and services that require real-time responses. Investors and developers will need to watch whether this efficiency translates into wider adoption and disruptive market shifts, or remains a specialized advancement.