Inception

What Inception does

Inception AI, Inc. is a Palo Alto-based AI company founded in 2024 by Stanford professor Stefano Ermon that develops diffusion-based large language models (dLLMs), an alternative architecture to traditional autoregressive LLMs. The company's Mercury product family generates text and code through parallel token refinement rather than sequential decoding, achieving approximately 1,000 tokens per second on NVIDIA Blackwell GPUs—roughly 10x faster than leading autoregressive models—while maintaining competitive benchmark quality (e.g., AIME 2025: 91.1%, GPQA: 73.6%) at substantially lower inference cost. The portfolio spans Mercury 2 (reasoning model with 128K context), Mercury Edit 2 (next-edit prediction for code), Mercury Coder (code generation), and the general Mercury chat model, with a research-stage multimodal extension called LaViDa. The team includes researchers from Stanford, UCLA, Cornell, Google DeepMind, Meta AI, Microsoft AI, and OpenAI, and the founders are credited with inventing foundational techniques including diffusion models, Flash Attention, and Direct Preference Optimization.

The product is distributed through an OpenAI-compatible REST API, the Mercury Chat playground, AWS Bedrock, Azure AI Foundry (with SOC2 and HIPAA compliance), OpenRouter, and Models.dev, with deep integrations into developer tooling (Zed, Continue, ProxyAI, Augment Code) and the Microsoft NLWeb open project. Inception generates revenue primarily through usage-based API pricing ($0.25 per 1M input tokens, $0.75 per 1M output tokens) supplemented by enterprise licensing on cloud marketplaces and custom contracts with dedicated capacity and SLAs. A 10-million-token free tier supports a self-serve developer acquisition motion running in parallel with a dedicated enterprise sales motion. Customer logos span coding tools (Zed, Augment Code, Buildglare, ProxyAI), enterprise search (SearchBlox), AI agent platforms (Radient, Skyvern), and real-time voice AI (Happyverse AI, OpenCall), with strategic partnerships extending into financial services (Bain & Company, Kensho), Arabic-language AI (Cerebras, MBZUAI, Jais 2), and government/security (Brain Co., Mirror Security/G42).

In November 2025, Inception raised a $50 million seed round led by Menlo Ventures with participation from Microsoft M12, NVentures, Snowflake Ventures, Databricks Ventures, Mayfield, and Innovation Endeavors, plus angels Andrew Ng and Andrej Karpathy. As of mid-2026, the company was reportedly in acquisition discussions with Microsoft at a valuation exceeding $1 billion.

Name

Inception

Legal name

Inception AI, Inc.

Website

https://inceptionlabs.ai

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

11–50 employees

Short description

Inception develops diffusion-based large language models (Mercury family) that generate tokens in parallel for ~10x faster inference than autoregressive LLMs, serving developers, voice AI platforms, enterprise search, and agent platforms via API and cloud marketplace channels.

Ownership category

akta.pro rank

Product category

AI Foundation Models

NAICS

Software Publishers (5132), Software Publishers (513210), Computer Systems Design and Related Services (5415)

SIC

Services-Prepackaged Software (7372)

akta.pro primary industry

Multimodal Generation (Text-Image-Video-Audio) & Creative Toolchains (HDAAACAG)

akta.pro secondary industries

LLMOps & Generative AI Platforms (Prompt/Agent Orchestration, RAG) (HDAEANAD), Fine-Tuning, Adaptation & Custom Model Training (PEFT/LoRA/RLHF) (HDAAACAC), Machine Translation & Multilingual NLP (HDAAADAD), MLOps/LLMOps & Model Lifecycle Management Services (BPAEAHAH)

Keywords

• Diffusion language models
• Generative AI foundation models
• AI inference API
• Code generation AI
• Reasoning LLM platform

Headquarters

HQ city

Palo Alto

HQ country

United States

HQ region

North America

Offices1 record

Markets served

GTM type

B2B

Offering type

Software

Revenue model

1. API Usage-Based Pricing: Pay-per-token pricing model for Mercury and Mercury Coder models through Inception API platform, with differentiated pricing for input tokens, output tokens, and cached inputs
2. Enterprise Licensing: Enterprise deployment licensing through AWS Bedrock and Azure AI Foundry with software license costs (e.g., $0.78/hour for Mercury on Azure) plus cloud compute costs
3. Custom Enterprise Contracts: Custom rate limits, SLA guarantees, volume-based pricing, dedicated capacity, and custom terms for large enterprise deployments

Pricing tiers

Go-to-market motion3 records

Distribution channels5 records

Marketing channels9 records

Core offering

Inception builds and deploys diffusion-based large language models (dLLMs) under the Mercury family that generate text and code through parallel iterative refinement rather than sequential autoregressive decoding. Their models deliver 5-10x faster inference at roughly half the cost of conventional LLMs while maintaining frontier-class quality, with distribution through a self-serve OpenAI-compatible API and enterprise deployments on AWS Bedrock and Azure AI Foundry.

Product overview

Inception is an AI research and product company developing diffusion-based large language models (dLLMs) that generate text and code through parallel refinement rather than sequential token-by-token generation. The product portfolio centers on the Mercury family of diffusion LLMs: Mercury 2 serves as the flagship reasoning model with 128K context, Mercury Edit 2 provides next-edit prediction for coding workflows, Mercury Coder powers code generation and apply-edit capabilities, and the general chat Mercury model enables conversational AI. The Inception API (OpenAI-compatible) and Mercury Chat playground provide developer access, while enterprise deployment options include AWS Bedrock, Azure AI Foundry, and model routers like OpenRouter. The company also partners with Bain & Company, Kensho, and Brain Co. for enterprise solutions, and collaborates on specialty models like Jais 2 (Arabic LLM with Cerebras/MBZUAI). Mercury's diffusion approach delivers 5-10x faster inference than traditional autoregressive models while maintaining competitive quality, enabling use cases in coding assistants, real-time voice agents, agentic loops, and enterprise search.

Differentiator

Problem solved

Functional benefit

Products and services

• Mercury 2 The flagship diffusion-based reasoning LLM from Inception achieving approximately 1,009 tokens per second on NVIDIA Blackwell GPUs through parallel token generation. Features 128K context window, tunable reasoning, native tool use, and schema-aligned JSON output for developers and enterprises building latency-sensitive AI applications.
• Mercury Edit 2 A small, coding-focused diffusion LLM optimized for next-edit prediction in coding workflows. Uses recent edits and codebase context to predict what developers will change next, delivering fast autocomplete and apply-edit suggestions.
• Mercury Coder The first commercial-scale diffusion LLM optimized for code generation, achieving 1000+ tokens per second on NVIDIA H100s. Supports fill-in-the-middle (FIM) and apply-edit capabilities for coding workflows.
• Mercury Coder Small A smaller coding-focused diffusion model variant available via API, running more than 5x faster than speed-optimized frontier models like GPT-4o Mini and Claude 3.5 Haiku while matching them in quality. Supports 32K context window and fill-in-the-middle workflows.
• Mercury (General Chat) First general chat diffusion LLM that matches the performance of speed-optimized frontier models like GPT-4.1 Nano and Claude 3.5 Haiku while running over 7x faster. Powers conversational AI applications, enterprise search, and voice interfaces.
• Inception API Platform REST API providing programmatic access to Mercury diffusion LLMs. OpenAI API compatible with support for chat completions, fill-in-the-middle, and apply-edit endpoints. New accounts receive 10 million free tokens.
• Mercury Chat Playground Interactive web-based playground for testing Mercury models. Allows developers to explore model capabilities and experiment with diffusion-based text generation.
• Mercury on Azure AI Foundry

Quantifiable outcome

• 1,000 tokens/second throughput vs ~71-89 tokens/second for GPT-5 Mini and Claude 4.5 Haiku
• +5 more outcomes

Named customers9 records

Segments5 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

Integration12 records

AI capability9 records

Feature5 records

Partnerships

17 partnerships are on record, tiered flagship, core and strategic.

• Augment CodeflagshipTechnology or Integration · 12 May 2026Production deployment of Mercury 2 for context compaction, model routing, and tool search; achieved 82% latency reduction and 90% cost savings
• ZedcoreTechnology or Integration · 30 March 2026Mercury Edit 2 integrated into Zed's edit prediction system for fast code editing
• SearchBloxflagshipTechnology or Integration · 12 January 2026Mercury-powered SearchAI for sub-second GenAI search with 60-90% lower inference costs across enterprise workloads
• Cerebras SystemsflagshipStrategic or Co-development Partner · 9 December 2025Co-developed Jais 2 Arabic LLM (70B parameters) trained on largest Arabic-first dataset; Cerebras provides wafer-scale computing infrastructure for training
• MBZUAI (Mohamed bin Zayed University of Artificial Intelligence)flagshipStrategic or Co-development Partner · 9 December 2025Co-developed Jais 2 Arabic LLM for high fluency, cultural depth, and safety in Arabic language AI
• Mirror SecuritystrategicStrategic or Co-development Partner · 24 November 2025Strategic partnership with G42's Inception to co-develop AI security products for government and enterprise clients
• Microsoft Azure AI FoundryflagshipTechnology or Integration · 18 November 2025Mercury available on Azure AI Foundry with enterprise-grade infrastructure, compliance standards (SOC2, HIPAA), and $0.78/hour software license
• ProxyAIcoreTechnology or Integration · 27 October 2025Partnership to bring faster, more consistent code edits to ProxyAI platform
• KenshostrategicImplementation/ SI/ Consulting Partner · 20 October 2025Strategic collaboration to develop AI solutions for financial institutions at GITEX Global 2025
• Bain & CompanystrategicImplementation/ SI/ Consulting Partner · 14 October 2025Strategic collaboration to develop and deploy AI solutions for financial services, starting with (In)Alpha for investment workflows
• Brain Co.strategicStrategic or Co-development Partner · 14 October 2025Strategic partnership to develop next-generation AI enterprise applications for public services, healthcare, and energy sectors
• G42 (Inception subsidiary context)coreOthers · 13 October 2025Inception is a G42 company with access to G42's AI infrastructure and regional presence in Middle East
• Amazon (AWS Bedrock & SageMaker JumpStart)flagshipTechnology or Integration · 27 August 2025Mercury and Mercury Coder available on Amazon Bedrock Marketplace and SageMaker JumpStart for enterprise deployment
• Continue (VS Code Extension)coreTechnology or Integration · 27 August 2025IDE integration partnership with Continue extension for Mercury Coder autocomplete and next-edit features
• RadientcoreTechnology or Integration · 19 August 2025Mercury powering model routing for Radient Automatic, enabling sub-second classification and 70% cost savings
• BuildglarecoreTechnology or Integration · 6 July 2025Mercury Coder powering Apply-Edit functionality for low-code web development platform
• Microsoft NLWebflagshipTechnology or Integration · 15 May 2025Founding LLM partner for Microsoft's NLWeb open project, enabling natural language interfaces for websites with ultra-fast Mercury dLLM

Scale indicators6 records

Recent moves6 records

Expansion highlights6 records

Direct peers

• OpenAI: Frontier foundation model provider (GPT-4o, GPT-5 family) competing directly with Mercury's reasoning and coding capabilities. Inception positions Mercury 2 as '2x faster than GPT-5.2' per customer testimonials, and offers an OpenAI-compatible API as a drop-in alternative.
• Anthropic: Enterprise-focused foundation model provider (Claude family). Anthropic's Claude Haiku 4.5 is benchmarked head-to-head with Mercury 2 on speed, and Inception's enterprise GTM via AWS Bedrock and Azure Foundry competes directly with Anthropic's distribution footprint.
• Mistral AI: European foundation model lab offering open and commercial LLMs with emphasis on efficiency and inference cost — a similar value proposition to Inception's Mercury dLLMs targeting latency-sensitive and cost-sensitive enterprise workloads.
• Cohere: Enterprise-focused foundation model provider offering API-accessible LLMs with private deployment and RAG tooling. Competes with Inception for enterprise search, retrieval, and agentic use cases through similar cloud marketplace channels.
• AI21 Labs: Foundation model provider offering task-specific and general LLMs (Jamba family) with enterprise focus. Similar stage and enterprise GTM motion to Inception, with comparable emphasis on efficiency and developer-friendly APIs.

Broad incumbents

• Google DeepMind: Develops Gemini family of foundation models with broad modality support (text, image, video, audio). A broad incumbent whose research output (including diffusion-based image/video models) competes architecturally with Inception's diffusion-based language approach.

Emerging players

• Groq: Provides ultra-low-latency LLM inference through custom LPU hardware. Competes with Inception on the same 'speed as a differentiator' thesis — Groq at the hardware/serving layer, Inception at the model architecture layer.
• Together AI: Open-source-focused AI inference and fine-tuning platform hosting multiple foundation models. Competes with Inception's API platform for developer mindshare in serving fast, cost-efficient LLM inference.
• Reka AI: Multimodal foundation model startup offering Reka Core, Flash, and Edge models. Comparable in stage and ambition to Inception, with overlapping multimodal and enterprise positioning.

Others

• Cerebras Systems: Wafer-scale AI compute provider and co-developer of Inception's Jais 2 Arabic LLM. While not a direct model competitor, Cerebras is an infrastructure partner and adjacent player in the speed-optimized AI stack that underpins Mercury's value proposition.

Market position

Strengths4 records

Weaknesses5 records

Competitive moat5 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/inceptionAILabs
• LinkedInlinkedin.com/company/inception-labs-ai

Compliance2 records

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds3 records

Investors10 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Inception