# Mercury 2: The First Reasoning Diffusion Language Model (1,000+ tokens/sec)

Source: https://www.youtube.com/watch?v=quOe8V2n9rU
Recap page: https://rapidrecap.app/video/quOe8V2n9rU
Generated: 2026-02-24T18:07:18.823+00:00

---
## Quick Overview

Inception Labs released Mercury 2, the first reasoning Diffusion Large Language Model (dLLM), which achieves over 1,000 tokens per second inference speed by generating tokens in parallel using a diffusion-based approach, significantly outpacing traditional autoregressive LLMs which require sequential, left-to-right generation.

**Key Points:**
- Mercury 2 is the first commercial-grade diffusion LLM, utilizing a diffusion-based approach for language generation, enabling parallel token generation.
- The model achieves over 1,000 tokens per second (t/s) inference speed, demonstrated to be significantly faster than autoregressive models like ChatGPT, which completed a 75-iteration task in 75 iterations versus Mercury's 14 iterations.
- The diffusion approach starts with noise and iteratively refines the output in parallel over multiple passes, contrasting with autoregressive models that generate one token sequentially.
- Mercury 2 demonstrates competitive reasoning quality, scoring 91.1 on AME 2025 and 73.6 on GPQA, while maintaining its speed advantage.
- The input cost for Mercury 2 is $0.25 per 1M tokens, and the output cost is $0.75 per 1M tokens, positioning it as a cost-effective option compared to other top-tier models.
- Inception Labs also offers Mercury Edit, a smaller, coding-focused dLLM optimized for low-latency coding tasks.

![Screenshot at 00:06: A side-by-side comparison demonstrating the speed difference between Autoregressive LLM \(left, 13 iterations completed\) and Inception Diffusion LLM \(right, 13 iterations completed\) when writing a function for LLM inference, visually highlighting the efficiency of the diffusion approach.](https://ss.rapidrecap.app/screens/quOe8V2n9rU/00-00-06.jpg)

**Context:** Inception Labs announced the release of Mercury 2, positioning it as a significant advancement in Large Language Models (LLMs) by applying diffusion-based techniques, traditionally used in image and video generation, to text generation. This marks a departure from the standard autoregressive method used by models like GPT-3 and LLaMA, aiming to deliver superior speed and efficiency while maintaining high reasoning quality.

## Detailed Analysis

Inception Labs introduced Mercury 2, the first reasoning Diffusion Large Language Model (dLLM), which leverages a diffusion-based approach to achieve unprecedented speed in text generation, reaching over 1,000 tokens per second (t/s). This contrasts sharply with traditional autoregressive LLMs (like ChatGPT shown in a benchmark), which generate text token by token sequentially, resulting in significantly higher iteration counts for the same task (e.g., 75 iterations for autoregressive vs. 14 for Mercury 2). The speed advantage is inherent to the model architecture, allowing parallel token generation across multiple refinement passes starting from noise, rather than relying on better hardware or infrastructure optimization alone. Benchmarks show Mercury 2 maintains competitive intelligence, achieving 91.1 on AME 2025 and 73.6 on GPQA, while offering favorable pricing ($0.25 input/ $0.75 output per 1M tokens). The company also details Mercury Edit, a coding-focused dLLM, and emphasizes that their foundational technology is applicable beyond text to image and video generation. The platform provides an OpenAI-compatible API, supporting integration with existing developer tools and workflows.

### Mercury 2 Introduction

- A new frontier in LLM speed
- Inception's breakthrough diffusion-based approach enables world's fastest, most efficient AI models with best-in-class quality
- Mercury 2 is the fastest reasoning LLM, ideal for complex applications where performance and speed are crucial.

### Speed Comparison (Diffusion vs. Autoregressive)

- Autoregressive LLM required 75 iterations to complete a task, while Inception Diffusion LLM required only 14 iterations (00:17). Mercury 2 achieves 1009 tokens/sec vs. Claude Hiku 4.5 at 88 t/s and GPT-3.5-mini at 71 t/s (01:50).

### Key Capabilities

- Speed and efficiency via parallel token generation
- Fine-grained control over output semantics
- Multimodal integration for handling text, images, audio, and video.

### Mercury 2 Benchmarks

- Scores 91.1 on AME 2025 and 73.6 on GPQA (Reasoning); 71.3 on MMLU (Instruction Following); 67.3 on LiveCodeBench (Code) (02:03).

### Pricing

- Input cost $0.25/1M tokens, Output cost $0.75/1M tokens for Mercury 2 (04:16).

### Applications

- Lightning fast agents for complex coding, Real-time voice engagement, Instant code editing, and Fast, creative co-pilots (07:15).

![Screenshot at 00:00: The landing page for Inception Labs introducing Mercury 2 as a new frontier in LLM speed using a diffusion-based approach.](https://ss.rapidrecap.app/screens/quOe8V2n9rU/00-00-00.jpg)
![Screenshot at 00:12: A split-screen comparison showing the Autoregressive LLM taking 75 iterations to complete a task while the Inception Diffusion LLM completes it in 14 iterations.](https://ss.rapidrecap.app/screens/quOe8V2n9rU/00-00-12.jpg)
![Screenshot at 01:50: A speed benchmark chart showing Mercury 2 achieving 1009 tokens/sec, significantly faster than competitors like Claude Hiku 4.5 \(88 t/s\) and GPT-3.5-mini \(71 t/s\).](https://ss.rapidrecap.app/screens/quOe8V2n9rU/00-01-50.jpg)
![Screenshot at 02:03: A bar chart detailing Mercury 2 benchmarks across Reasoning \(AME 2025: 91.1, GPQA: 73.6\), Instruction Following, and Code metrics.](https://ss.rapidrecap.app/screens/quOe8V2n9rU/00-02-03.jpg)
![Screenshot at 07:15: A section highlighting the four core application areas for Mercury: Lightning fast agents, Real-time voice, Instant code editing, and Fast, creative co-pilots.](https://ss.rapidrecap.app/screens/quOe8V2n9rU/00-07-15.jpg)