Inception Launches Mercury 3 Diffusion LLM, Claiming 8x Speed Over Luna and Haiku

Inception has officially launched Mercury 3, a new diffusion-based large language model that the company claims delivers superior speed and benchmark performance compared to established rivals. According to reporting by Techmeme, the debut positions the model as a high-speed alternative to current market leaders in the agentic AI sector.

The vendor states that Mercury 3 generates roughly 1,068 tokens per second, a metric that places it significantly ahead of competitors in self-reported testing. Jason Hiner of The Deep View noted that Inception asserts the model tops GPT-6 Luna, Claude Haiku 5.5, and GLM-5.3 Flash on specific agentic benchmarks while operating at four to eight times the speed.

Context and Background

As Threat Times reported on October 20, 2026, Inception introduced Mercury 3 with claims of superior speed and benchmark performance over rivals GPT-6 Luna and Claude Haiku 5.5 in self-reported tests. The initial coverage highlighted the model's generation rate of approximately 1,000 tokens per second, positioning it as a high-speed alternative to established models like Luna and Haiku.

The technology sector has traditionally invested heavily in standard large language models and the massive data centers required to scale them. Inception represents one of the newer laboratories betting on diffusion-based architectures to challenge this dominant paradigm. The company’s approach suggests a shift toward models that can deliver faster inference times without relying solely on the massive infrastructure associated with traditional scaling methods.

Who Is Affected

Developers and enterprises evaluating agentic AI solutions may find Mercury 3 relevant for workflows requiring rapid token generation. The performance claims specifically target comparisons with GPT-6 Luna, Claude Haiku 5.5, and GLM-5.3 Flash, indicating that users of these platforms might consider the new model for latency-sensitive applications.

The announcement primarily impacts the competitive landscape for high-speed AI inference. By asserting dominance on agentic benchmarks, Inception aims to attract users who prioritize execution speed alongside task completion accuracy. The vendor’s self-reported metrics suggest a potential disruption in the market for fast-processing language models, though independent verification of these specific speed and performance claims remains pending.

What should you do?

Organizations should monitor independent benchmarks to verify the self-reported speed and performance claims made by Inception. Security and technology professionals should assess whether the diffusion-based architecture introduces new attack surfaces or data handling requirements compared to traditional transformer models. Evaluate current AI infrastructure dependencies to determine if a switch to a faster model aligns with existing security and compliance protocols.

Background: Large language models

Large language models are a specific type of neural network trained on text. Generative AI is the broader umbrella term for any system that creates new content, including images and code. Not all generative AI uses large language models, but most modern text generators do.

Read the full guide: Large Language Models vs Generative AI: The Distinction and Security Implications

What to do and how to stay safe: Inception

• Review vendor documentation to understand the specific agentic benchmarks used for performance comparisons.
• Assess data privacy implications of integrating new third-party language models into sensitive workflows.
• Monitor for independent third-party testing results that validate the claimed speed improvements.
• Evaluate current AI provider contracts to determine if switching costs justify potential performance gains.

Step-by-step guide: Prompt Injection Warning Signs: How to Spot AI Manipulation

General security guidance from the Threat Times newsroom. It is not confirmed advice from the organisations named in this story.

Frequently asked questions

How fast does Inception claim Mercury 3 is?

The vendor claims the model generates roughly 1,068 tokens per second, which is four to eight times faster than competitors.

Which models does Mercury 3 outperform according to Inception?

Inception states that Mercury 3 tops GPT-6 Luna, Claude Haiku 5.5, and GLM-5.3 Flash on agentic benchmarks.

Is Mercury 3 a traditional transformer model?

No, Mercury 3 is described as a diffusion-based large language model, distinguishing it from standard architectures.