Mercury 2.5 is the largest diffusion language model trained to date and, per Inception, the fastest reasoning LLM in production. Instead of emitting tokens one at a time, it produces and refines many tokens in parallel — the core idea behind diffusion-based language models.
Inception reports a 40% increase in intelligence over Mercury 2 and 1,107 tokens per second on widely available NVIDIA GPUs, placing it alongside cost-optimised frontier models on quality while running considerably faster.
Latency-sensitive agents
Throughput above 1,000 tokens/sec makes multi-step agent loops feel interactive rather than batched.
High-volume generation
Parallel decoding keeps cost per token low at scale.
Released by Inception on August 31, 2026 and available on AI/ML API since September 8, 2026.
Measured on the AI/ML API gateway across live traffic. Figures are medians unless stated. Last updated 2026-10-12.
Mercury 2.5 is documented for fast chat, agents, and structured workflows. It is also suited to search and RAG pipelines that make many calls per request.
It is documented for voice agents and voice pipelines. Inception also lists Mercury Voice use cases including customer support, patient care, education, and gaming, though those uses are shown for the voice product line rather than Mercury 2.5 specifically.
It is documented for coding-assistant tasks, including context compaction, routing, autocomplete, and next-edit suggestions. Inception also says it serves latency-sensitive coding products and coding subagents.
import requests r = requests.post( "https://api.aimlapi.com/v1/chat/completions", headers={"Authorization": "Bearer " + AIMLAPI_KEY}, json={ "model": "inception/mercury-2.5", "messages": [ { "role": "user", "content": "Hello!" } ] }, ) print(r.json())
const r = await fetch("https://api.aimlapi.com/v1/chat/completions", { method: "POST", headers: { Authorization: `Bearer ${process.env.AIMLAPI_KEY}`, "Content-Type": "application/json", }, body: JSON.stringify({ "model": "inception/mercury-2.5", "messages": [ { "role": "user", "content": "Hello!" } ] }), }); console.log(await r.json());
curl -X POST https://api.aimlapi.com/v1/chat/completions \ -H "Authorization: Bearer $AIMLAPI_KEY" \ -H "Content-Type: application/json" \ -d '{"model":"inception/mercury-2.5","messages":[{"role":"user","content":"Hello!"}]}'
OpenAI-compatible — swap the base URL and it works with your existing SDK.
| Type | Price |
|---|---|
| Input | |
| Output | |
| Cached input |
| Model | Input | Output | Context | Best for |
|---|---|---|---|---|
Mercury 2.5 This page | Low-latency reasoning and agent loops | |||
| Reasoning + agents | ||||
| Reasoning + agents | ||||
| Balanced coding + agents | ||||
| Reasoning + agents |
Mercury 2.5 is a diffusion language model from Inception. Unlike models that emit one token after another, it generates and refines many tokens in parallel.
A standard LLM predicts tokens one at a time, each conditioned on the previous one. A diffusion language model starts from a rough draft of the whole output and refines it over several passes, so many tokens improve at once.
Inception reports throughput of 1,107 tokens per second on widely available NVIDIA GPUs, and describes it as the fastest reasoning model in production.
Yes. It supports function calling and can issue parallel tool calls within a single turn.
Yes. It supports schema-aligned JSON output, so responses can be constrained to the shape your code expects.
Yes. Reasoning depth is tunable, so you can trade latency against deliberation on a per-request basis.
Mercury 2.5 takes text and image input and returns text. It also supports file input and web search.
Send a request to the chat completions endpoint with the model id inception/mercury-2.5. The request shape is the same as for any other chat model on AI/ML API.
Inception reports roughly a 40% increase in intelligence over Mercury 2, while keeping the parallel decoding that makes the family fast.
Latency-sensitive work such as agent loops that make many sequential calls, and high-volume generation where throughput is the constraint.
