Mercury 2

Mercury 2 represents a departure from traditional autoregressive language models by utilizing a diffusion-based architecture. While standard models generate text sequentially, token by token, this model functions through parallel refinement, allowing it to produce and revise multiple tokens simultaneously. This design intent focuses on solving the latency bottlenecks inherent in complex production environments, such as agent loops, real-time voice interactions, and automated coding tasks. By acting more like an editor refining a full draft rather than a typewriter, the model achieves significant throughput gains while maintaining reasoning-grade quality.

Developed by researchers from Stanford, UCLA, and Cornell, the model leverages the team's foundational work in diffusion to bring high-performance inference to real-world applications. By moving away from sequential decoding, the model achieves speeds exceeding 1,000 tokens per second on standard hardware, offering a distinct advantage for workflows where latency compounds across multiple steps. This architectural shift positions the model as a practical solution for high-value production deployments that require both rapid response times and the ability to handle complex, multi-step reasoning tasks at a lower cost.

OpenRouterinception/mercury-2mercury

Open provider page Provider docs

Quick Info

Provider: OpenRouter
Model key: inception/mercury-2
Release date: Mar 4, 2026
Last updated: Mar 4, 2026
Input modalities
Output modalities
Capabilities

Cost

Input token cost: $0.25
Output token cost: $0.75

Limits

Output tokens: 50,000 tokens
Context window: 128,000 tokens

Transparent token rates

Compare Mercury 2 pricing

Rates are shown per one million tokens. Combined means one million input plus one million output tokens.

Browse this family

Latest news about Mercury 2

Vercel AI Gateway

Official sourceAnnouncement

A second Inception blog post by Joyce Chen positions Mercury 2 as "the first reasoning model fast enough to pick up the phone," targeting real-time voice applications. It argues that GPT 5.x, Claude Sonnet, and Gemini Pro fail to meet sub-500ms latency requirements for voice agents when running reasoning chain-of-thoug The post contextualizes Mercury 2's launch against the imminent sunsetting of GPT-4.1, which the author identifies as the de facto default model for most production voice agents as of April 2025 due to its balance of instruction-following, tool-calling, and realtime latency. Inception positions Mercury 2 as the first r

Inception

Official sourceAnnouncement

Inception published a first-party blog post officially launching Mercury 2, the company's diffusion-based reasoning large language model. The post, authored by CEO Stefano Ermon, introduces Mercury 2 as the fastest reasoning LLM, built on parallel-refinement diffusion decoding rather than token-by-token autoregressive The blog frames Mercury 2's release as a meaningful step in moving diffusion LLM technology from research into production, arguing that diffusion's parallel-token generation fundamentally changes the intelligence-versus-latency trade-off by enabling reasoning-grade quality within real-time latency budgets. Compared wit

OpenRouter

CoverageBenchmark

Mercury 2 targets structured tasks with schema-aligned JSON output; supports OpenAI API drop-in integration, for simpler deployment.

OpenRouter

Coverage

Inception Labs introduces Mercury 2, a diffusion-based LLM designed for high-speed, multi-step reasoning tasks with a 128K context window.

OpenRouter

Coverage

Inception's Mercury 2 diffusion LLM hits 1,196 tokens/sec at $0.25/M input, Meta signs $100B+ AMD compute deal, MatX raises $500M to challenge Nvidia's Rubin Ultra. (166 chars)

OpenRouter

Coverage

The news blog specialized in Japanese culture, odd news, gadgets and all other funny stuffs. Updated everyday.

Inception

CoverageBenchmark

Artificial Analysis has published an independent intelligence, performance, and price profile for Inception's Mercury 2, a proprietary model released in February 2026. The third-party evaluation scores Mercury 2 at 15 on the Artificial Analysis Intelligence Index v4.2 — above the comparable-model median of 12 — and mea The benchmark page classifies Mercury 2 as a reasoning-capable model (noting a non-reasoning variant may also exist) with text-in/text-out modalities and a 128K token context window. Notably, Artificial Analysis's measured speed of 794.5 tokens per second is lower than Inception's vendor-reported 1,009 tokens per secon

OpenRouter

Official sourceBenchmark

Benchmark scores and performance metrics for Inception: Mercury 2 - Mercury 2 is an extremely fast reasoning LLM, and the first reasoning diffusion LLM (dLLM). Instead of generating tokens sequentially, Mercury 2 produces and refines multiple tokens in parallel, achieving >1,000 tokens/sec on standard GPUs. Mercury 2 i