New AI model claims to be 8x faster than GPT-6
What's the story
Inception, a California-based AI research and product firm, has launched its latest diffusion language model, Mercury 3. The company claims that the new model outperforms speed-optimized competitors such as GPT-6 Luna and Claude Haiku 5.5 on various benchmarks while running four to eight times faster. Mercury 3 can generate over 1,000 tokens per second on widely available NVIDIA GPUs and supports a context window of one million tokens.
Model evolution
Mercury 3 is 40 times larger than Mercury 1
Inception claims that Mercury 3 is more than 40 times larger than its first-generation model, Mercury 1, while maintaining similar processing speeds.
The company says this shows that diffusion-based language models can be made more capable without compromising the speed benefits of their underlying architecture.
This is a major leap forward in the world of AI language models.
Speed advantage
Diffusion language models generate and refine multiple tokens in parallel
Unlike traditional large language models that generate text sequentially, diffusion language models generate and refine multiple tokens in parallel.
This makes them faster, potentially cutting down on delays in applications requiring frequent interactions with a model.
However, developers have had to choose between model capability and response speed until now.
Mercury 3 solves this problem by delivering performance comparable to or better than OpenAI and Anthropic models while maintaining inference speeds above 1,000 tokens per second.
Enhanced features
Context window of 1 million tokens
Mercury 3 is the fourth model from Inception since it launched Mercury 1 in June 2025.
The new model nearly quadruples the amount of information it can handle at once, with a context window capacity of one million tokens.
Inception is targeting applications where response delays can affect usability, including interactive tutoring, live translation, writing assistants, and automated software agents.
Usability impact
Early-access customers are already testing the model in production workflows
In agent-based systems, models often perform multiple sequential tasks. Delays at each stage can increase the time required to complete an operation.
Inception says Mercury 3's faster inference could help reduce these cumulative delays without requiring developers to switch to less capable models.
Early-access customers are already testing the model in production workflows.