Mercury 3

Name: Mercury 3
Author: Inception Labs

Released

2026-10-19

Last refreshed

2026-10-19

Status

Researched 1d ago

ProprietaryCommercial use: restrictedRAGAgentsLong contextClassificationJSON / Tool use

Inception Labs releases · 3 in the last 12 monthsChangelog →

Specifications

Family: Mercury
Released: 2026-10-19
Context: 1M
Max output: 128K
Openness: Proprietary
License: ProprietaryRestricted
Weights: Not released
Code: Unknown

Created by

Inception Labs

Inception Labs is an AI company building diffusion-based large language models (dLLMs), a fundamentally different archit

San Francisco, CA

Founded 2024

Website

Pricing

Output / 1M

$0.350

Input / 1M

$0.100

Cheapest of 3 routes · OpenRouter · cache read $0.010

Providers(3)

Inception OpenRouter Vercel AI Gateway

View 3 provider routes

Links

Website

About

Mercury 3 is Inception's diffusion LLM (dLLM), released 2026-10-19 on the Inception API (model id mercury-3) and through OpenRouter, Baseten and Vercel. It has a 1M-token context window and up to 131,072 output tokens, takes text in and returns text, and supports tool calling and structured outputs via the OpenAI-compatible v1/chat/completions endpoint. Reasoning effort can be set to instant, low, medium or high, and temperature from 0.5 to 1.0. Inception reports 1,068 output tokens per second on NVIDIA GPUs and says the model is more than 40x larger in total parameters than Mercury 1 (absolute size not disclosed).

Top use-case fit: coding, agents, and build tasks

Agents

Q/$ A

Providers(3)

Compare all 3

Compare API pricing across 3 tracked provider routes for input and output tokens, batch, and cached reads when available.

Provider	Input / 1M	Output / 1M	Cache	Route
OpenRouter	$0.100	$0.350	read $0.010	Serverless
Vercel AI Gateway	$0.100	$0.350	read $0.010	Serverless
Inception	$0.200	$0.700	read $0.020	Serverless

Capabilities

ReasoningJSON / Tool useStructured OutputsPrompt Caching

Benchmark peer barsfor Agents

MultiChallengeRank 8 of 16

Gemini 3.1 Pro Preview

71.4

69.2

67.6

63.8

60.4

Benchmark scores(2)

Scores are benchmark-specific and are direction-aware: the same numeric gap can mean very different outcomes across suites. Use the leaderboard context and this model's provider route to decide whether the winning margin is meaningful for your workload.

Benchmark	Score	Version	Evaluation	Source
Google-Proof Q&A	89.2	GPQA DiamondObserved 2026-10-19	Configuration: Mercury 3 (medium) Evaluator: Inception Confidence: self-reported Notes: Row 'GPQA Diamond': Mercury 3 (MEDIUM) 89.23 (GPT-6 Luna 86.36, Claude Haiku 5.5 84.51 in same table). Inception launch blog 'Mercury 3 — Benchmark results vs. models in its class' table image (https://framerusercontent.com/images/YO5OXn25zt3KJBJkUxO2kAwmS6o.png, evidence/blog-img-YO5OXn25.png; values transcribed by eye from printed table). Vendor-reported. Retrieved 2026-10-19.	Source
MultiChallenge	60.4	MultiChallengeObserved 2026-10-19	Configuration: Mercury 3 (medium) Evaluator: Inception Confidence: self-reported Notes: Row 'MultiChallenge': Mercury 3 (MEDIUM) 60.35. Inception launch blog 'Mercury 3 — Benchmark results vs. models in its class' table image (https://framerusercontent.com/images/YO5OXn25zt3KJBJkUxO2kAwmS6o.png, evidence/blog-img-YO5OXn25.png; values transcribed by eye from printed table). Vendor-reported. Retrieved 2026-10-19.	Source