Inception Releases Mercury 3 Diffusion LLM, Claiming Speed Gains Over GPT-6 Luna
Inception says its new diffusion model generates over 1,000 tokens per second, outpacing major rivals in self-reported benchmarks.
Key points
- Inception released Mercury 3, a diffusion-based language model, on Monday.
- The company claims the model runs four to eight times faster than GPT-6 Luna and Claude Haiku 5.5.
- Launch pricing is temporarily halved to $0.10 per million input tokens and $0.35 per million output tokens.
Inception, an artificial intelligence laboratory based in Redwood City, released its Mercury 3 model on Monday. The company describes the system as a diffusion large language model, a design architecture that differs significantly from the autoregressive systems powering most mainstream chatbots and agents.
According to self-reported benchmarks provided by Inception, Mercury 3 outperforms OpenAI’s GPT-6 Luna, Anthropic’s Claude Haiku 5.5, and Zhipu’s GLM-5.3 Flash across various tests. The company states the model runs four to eight times faster than these competitors while generating more than 1,000 tokens per second on Nvidia hardware. No independent third-party group has published verification of these performance figures yet.
Architecture and Performance Claims
Most current language models are autoregressive, generating text one token at a time in a sequential chain. In contrast, Inception says Mercury 3 uses a diffusion approach similar to image generation tools. The model starts with rough, noisy output and refines multiple tokens in parallel until the text resolves. Inception attributes its speed advantage to this parallel processing method.
The new model is more than 40 times larger than Inception’s first version. The company claims it is about three times as capable on standard benchmarks while maintaining similar speeds. Mercury 3 features a context window of one million tokens, an increase from the 260,000-token limit in the previous version. This allows the system to process significantly larger amounts of text simultaneously.
Pricing and Development Methodology
Mercury 3 carries a base price of $0.20 per million input tokens and $0.70 per million output tokens. For the launch period, Inception is halving both rates to $0.10 and $0.35, respectively. The company has not specified how long this promotional pricing will remain active. This pricing structure undercuts rivals on output costs, continuing a broader industry trend of reducing fees.
Inception chief executive Stefano Ermon told The Deep View that the model’s speed makes it suitable for agents and coding tools that require rapid, repeated calls. Ermon stated the system provides answers almost instantly. The lab also revealed it used swarms of Mercury 3 agents to improve the model recursively, a technique that added approximately 16% to the speed gains achieved through manual engineering.
What should you do?
Security and technology professionals should monitor independent evaluations of diffusion-based models before integrating them into production environments. As Threat Times reported on October 20, 2026, Inception previously highlighted significant speed gains for this model series. Organizations relying on high-throughput AI agents should assess whether the claimed speed benefits justify potential risks associated with unverified performance metrics.
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: Inception
- Verify performance claims through independent benchmarking before deploying new models in critical workflows.
- Review vendor pricing terms to understand the duration of promotional rates and potential cost spikes.
- Assess the security implications of recursive self-improvement techniques used in model training.
- Monitor for third-party audits or security disclosures regarding the new diffusion architecture.
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 does Mercury 3 differ from standard language models?
Mercury 3 uses a diffusion approach that refines multiple tokens in parallel, unlike autoregressive models that generate text one token at a time.
What is the cost of using Mercury 3 during its launch?
Inception is offering launch pricing of $0.10 per million input tokens and $0.35 per million output tokens, half the standard rates.
Has an independent group verified Mercury 3's performance claims?
No, the performance figures are self-reported by Inception, and no outside group has published a test of the model yet.
