Inception launches Mercury 3, a diffusion-based LLM claiming speeds up to 8x faster than Luna and Haiku
The new model generates roughly 1,000 tokens per second, positioning it as a high-speed alternative to established models like Luna and Haiku.
Key points
- Inception has released Mercury 3, a large language model built on diffusion technology rather than traditional autoregressive methods.
- The company claims the new model performs comparably to Luna and Haiku while operating four to eight times faster.
- Mercury 3 achieves a generation speed of approximately 1,000 tokens per second, according to reporting by Jason Hiner at The Deep View.
Inception has officially launched Mercury 3, a new large language model that utilizes diffusion-based architecture. This release introduces a model that the company claims is comparable to Luna and Haiku models but significantly faster in generation speed.
According to Jason Hiner at The Deep View, Inception claims that Mercury 3 offers performance comparable to established models such as Luna and Haiku. The primary differentiator is speed, with the new model reportedly generating content four to eight times faster than these specific competitors. This performance metric is highlighted as a key advantage for users seeking rapid text generation capabilities.
Technical Specifications and Performance Claims
The core technical achievement of Mercury 3 is its ability to generate approximately 1,000 tokens per second. This throughput represents a substantial increase over the generation rates typically associated with models like Luna and Haiku. By leveraging diffusion-based techniques, Inception aims to reduce the latency often experienced in large language model interactions.
The diffusion-based approach differs from the autoregressive methods used by most major AI providers. While the industry has heavily invested in traditional large language models and the massive data centers required to scale them, Inception is testing an alternative pathway. This method potentially offers a more efficient route to high-speed inference.
Industry Context and Competitive Landscape
The launch of Mercury 3 occurs against a backdrop of intense competition in the generative AI sector. Most major technology firms have bet heavily on expanding their infrastructure to support larger and more complex models. Inception’s strategy contrasts with this trend by focusing on architectural efficiency rather than sheer scale.
By positioning Mercury 3 as comparable to Luna and Haiku, Inception is targeting the segment of the market that values speed and responsiveness. These reference models are known for their specific performance characteristics, making them suitable benchmarks for evaluating new entrants. The claim of being four to eight times faster suggests a significant
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.
What to do and how to stay safe
- Sudden role changes in output signal a successful override of system instructions.
- Hidden metadata in images or documents often carries invisible injection payloads.
- Immediate isolation of the affected session prevents lateral movement within the AI workflow.
Prompt injections exploit the model’s ability to follow instructions by hiding malicious commands in input data. Isolate affected sessions immediately and implement external guardrails to filter inputs before they reach the model.
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.
