← Back to articles

Mercury 2.5

2026-09-08 | original

read original get Hands-On Large Language Models (O'Reilly book) → more articles

Why This Matters

Inception's Mercury 2.5 shows diffusion-based language models maturing into viable production systems, claiming a 40% intelligence gain over Mercury 2 while holding ~1,107 tokens/sec on standard NVIDIA GPUs. That combination of frontier-lite quality at very low latency and price ($0.20/$0.75 per million tokens, 80% off at launch) matters for agentic workloads like search, voice, and coding where dozens of model calls must fit inside one user interaction.

Key Takeaways

Mercury 2.5 is billed as the largest and most capable diffusion LLM to date, with 40% higher intelligence than Mercury 2 and a 260K context window.
Performance is positioned against cost-optimized frontier models (GPT-5.6 Luna Low, Gemini 3.5 Flash-Lite, Claude Haiku 4.5) at aggressive pricing.
Training was guided by real production failure cases and customer feedback rather than benchmarks alone, targeting latency-sensitive search, voice, and coding pipelines.

Worth a Look

Hands-On Large Language Models (O'Reilly book) — If a launch like Mercury 2.5 makes you want to understand what's actually happening under the hood of diffusion and transformer language models, this O'Reilly guide is a great place to start. It's a visual, code-along introduction to tokens, embeddings, and generation that pairs well with actually building on a fast, low-cost API.

See Hands-On Large Language Models (O'Reilly book) on Amazon → Affiliate link — we may earn a commission on purchases, at no extra cost to you. Product picked by AI based on this article; it is not a tested recommendation.

Today, we’re releasing Mercury 2.5, our most capable production model yet. It is a significant step up in quality over Mercury 2, with the same low-latency, low-cost serving profile.

Since Mercury 2’s launch, thousands of developers have built with it, dozens of enterprises have put it into production, and usage has grown over an order of magnitude. It now serves latency-sensitive workloads across search, voice, and coding products.

Those workloads gave us a clearer signal than benchmarks alone. We used customer feedback and production failure cases to sharpen the evals and focus training. Mercury 2.5 is the first result of that loop.

What changed

Mercury 2.5 is the most capable diffusion LLM on the market. To our knowledge, it is the largest diffusion language model ever trained.

Quality: 40% increase in intelligence from Mercury 2. Comparable to cost-optimized frontier models like GPT-5.6 Luna (Low), Gemini 3.5 Flash-Lite, and Claude Haiku 4.5.

Speed: 1,107 tokens per second on widely-available NVIDIA GPUs.

Context: 260K tokens.

Price: $0.20 per million input and $0.75 per million output. At launch, Mercury 2.5 is 80% off at $0.04 per million input and $0.15 per million output.

Capabilities: Tunable reasoning, parallel tool calls, and schema-aligned JSON.

... continue reading

Explore topics: mercury 2.5 diffusion llm inception nvidia gpus llm inference pricing