What Reflection AI does

Reflection AI, Inc. is a New York-domiciled, venture-backed AI lab founded in 2024 by former Google DeepMind researchers Misha Laskin (CEO) and Ioannis Antonoglou (CTO). The company develops frontier open-weight foundation models — large language models including massive Mixture-of-Experts architectures — trained on a proprietary large-scale LLM and reinforcement learning platform that integrates pre-training and post-training pipelines from the ground up. Its stated mission is to keep superintelligence open, forkable, and accessible rather than concentrated inside a few closed labs, and its current product focus is an autonomous coding system that plans, writes, and refines software, intended as a stepping stone toward general agentic reasoning. The organization operates as a "small talent-dense team" of 101-250 employees with offices in San Francisco, New York, and London, drawing research staff from DeepMind, OpenAI, Google Brain, Meta, Character.AI, and Anthropic.

The commercial model combines open-weight model distribution with bilateral enterprise, government, and sovereign-AI deployments. The company sells quote-based, multi-year services and licensing arrangements to large enterprise and government customers — most prominently the U.S. Department of Defense (Impact Levels 6 and 7 classified networks), Shinsegae Group (a 250MW AI data center and retail AI joint venture in South Korea), and Dell Technologies (Dell AI Factory with NVIDIA ecosystem). Distribution is reinforced through the NVIDIA Nemotron Coalition and the Dell AI Ecosystem Program, while the underlying compute is sourced through hyperscale agreements such as the $6.3B SpaceX Colossus 2 commitment. Pricing is not publicly disclosed; revenue mechanics are negotiated bilaterally, and as of March 2026 the company has yet to generate meaningful recurring revenue despite having raised approximately $2B at an $8B valuation in October 2025, with an additional $2.5B reported in process at a $25B pre-money valuation.

Name

Reflection AI

Legal name

Reflection AI, Inc.

Website

https://www.reflection.ai

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

101–250 employees

Short description

Reflection AI builds frontier open-weight foundation models and a proprietary large-scale LLM and reinforcement learning training stack, with autonomous coding as its current flagship product, serving enterprises, governments, and developers seeking an open alternative to closed AI labs.

Ownership category

akta.pro rank

Product category

Foundation Models

NAICS

Software Publishers (5132)

SIC

Services-Prepackaged Software (7372)

akta.pro primary industry

Foundation Model Developers (LLM/Multimodal Model Labs) (HDAAACAA)

akta.pro secondary industries

Open-Source Model Ecosystems & Model Marketplaces (HDAAACAM), Model Development & Training Platforms (AutoML, Notebooks, Feature Stores) (HDAEANAB), Safety, Alignment & Content Moderation (Guardrails, Red Teaming) (HDAAACAI)

Keywords

• Frontier AI models
• Open foundation models
• Autonomous coding agents
• Large language models
• Reinforcement learning platforms

Headquarters

HQ city

New York

HQ country

United States

HQ region

North America

Offices4 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Marketing or Sales, Operations

Revenue model

1. Sustainable commercial model for open intelligence: Reflection has stated it has identified a "scalable commercial model that aligns with our open intelligence strategy, ensuring we can continue building and releasing frontier models sustainably." Specific revenue mechanics are not publicly disclosed, and analysts note the company has yet to generate meaningful revenue.
2. Enterprise and government AI deployments: Large-scale contracts with enterprise, government, and sovereign customers — including U.S. Department of Defense classified network deals, Shinsegae joint venture for Korea's largest AI data center, and Dell AI ecosystem integration — likely structured as multi-year service or licensing agreements.

Pricing tiers

Go-to-market motion1 record

Distribution channels5 records

Marketing channels7 records

Core offering

Reflection AI builds open-weight frontier foundation models, including large-scale Mixture-of-Experts language models, trained on a proprietary end-to-end pretraining and reinforcement learning platform. The company's current commercial flagship is an autonomous coding system that plans, writes, and refines complex software, with a stated plan to extend the same autonomous capabilities to broader computer-based knowledge work. It distributes capabilities to enterprises, governments, and nation states through direct sales, OEM partnerships (NVIDIA, Dell), and open-weight releases.

Product overview

Reflection AI is building a unified open intelligence platform with three interconnected components: (1) a proprietary large-scale LLM and reinforcement learning training platform that trains massive Mixture-of-Experts (MoEs) models at frontier scale, (2) open foundation models (frontier open-weight LLMs) released openly for individuals, agents, enterprises, and nation states, and (3) an autonomous coding system that is the company's current flagship product focus and the proving ground for extending these capabilities to all other categories of computer-based work. The training platform underpins the open foundation models, and the autonomous coding system represents the first applied use case, with a stated long-term plan to expand from coding to general agentic reasoning and broader computer-based work.

Differentiator

Problem solved

Functional benefit

Products and services

• Frontier Open Intelligence (Open Foundation Models) Open-weight frontier foundation models, including large Mixture-of-Experts (MoE) language models, trained on Reflection's proprietary large-scale LLM and reinforcement learning platform. Intended to be released openly for individuals, developers, enterprises, and nation states as a forkable, customizable, and embeddable alternative to closed frontier AI labs.
• Autonomous Coding System Superintelligent autonomous coding system that plans, writes, and refines complex software automatically. It is the company's current flagship applied product and the proving ground for extending autonomous agentic capabilities to broader categories of computer-based knowledge work.
• Large-Scale LLM and Reinforcement Learning Training Platform Proprietary training infrastructure that integrates large-scale pretraining with reinforcement learning from the ground up, capable of training massive Mixture-of-Experts models at frontier scale. Underpins Reflection's open foundation models and the autonomous coding system, and supports sovereign AI deployments (e.g., Shinsegae's planned 250MW South Korea AI data center) on Nvidia GB300 hardware.

Quantifiable outcome

• Raised $2 billion at an $8 billion valuation in October 2025 (a 15x valuation increase over ~20 months from $545M to $8B).
• +4 more outcomes

Named customers4 records

Ideal customer profiles2 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

AI capability7 records

Feature6 records

Partnerships

Six partnerships are on record, tiered core / flagship and major.

• SpaceX (xAI / Colossus 2)core / flagshipTechnology or Integration · 22 June 2026SpaceX signed a multibillion-dollar compute agreement with Reflection AI on June 22, 2026, granting immediate access to Nvidia GB300 AI chips at SpaceX's Colossus 2 data center in Memphis, Tennessee. Reflection will pay $150M/month starting July 1, 2026 through 2029, totaling approximately $6.3B. Either party can terminate with 90 days' notice after an initial three-month period.
• Dell TechnologiesmajorStrategic or Co-development Partner · 18 May 2026At Dell Technologies World 2026 (May 18, 2026), Dell named Reflection as a partner in the new Dell AI Ecosystem Program alongside Google Cloud, OpenAI, Palantir, Hugging Face, SpaceXAI, and ServiceNow; Dell's AI Factory with NVIDIA already serves over 5,000 customers.
• U.S. Department of Defense (Pentagon / Department of War)core / flagshipOthers · 1 May 2026On May 1, 2026 the Pentagon signed agreements with Reflection AI alongside SpaceX, OpenAI, Google, Nvidia, Microsoft, AWS, and Oracle to deploy AI capabilities on classified networks at Impact Levels 6 and 7, supporting the "AI-first fighting force" strategy. Reflection is the only dedicated open-weight AI vendor in the group and is explicitly framed as the "American alternative" to Chinese open-source models.
• NVIDIA Nemotron Coalition (Black Forest Labs, Cursor, LangChain, Mistral AI, Perplexity, Sarvam, Thinking Machines Lab)majorStrategic or Co-development Partner · 20 March 2026Reflection AI is a member of NVIDIA's Nemotron Coalition for frontier open models; coalition members share research, expertise, data, and compute on NVIDIA DGX Cloud infrastructure, with the first project being a base model co-developed by Mistral AI and NVIDIA underpinning the Nemotron 4 family.
• Shinsegae Groupcore / flagshipStrategic or Co-development Partner · 17 March 2026Shinsegae signed an MOU with Reflection AI in San Francisco on March 16, 2026 to build Korea's largest AI data center with up to 250MW capacity; the project is supported under the U.S. AI Exports Program with U.S. Secretary of Commerce Howard Lutnick attending. Shinsegae subsequently pivoted from OpenAI to Reflection AI for retail AI development in April 2026.
• U.S. Department of Commerce (American AI Exports Program)majorStrategic or Co-development Partner · 17 March 2026Reflection's Shinsegae partnership (Korea's largest AI data center, 250MW) is the first flagship outcome of the U.S. AI Exports Program, with U.S. Secretary of Commerce Howard Lutnick attending the MOU signing; Pax Silica Fund cited the Shinsegae deal as a model for alliance-based AI sovereignty.

Scale indicators10 records

Recent moves7 records

Expansion highlights6 records

Direct peers

• Mistral AI: Mistral AI is the leading European open-weight foundation model developer, releasing frontier LLMs (Mixtral, Mistral Large) under permissive licenses. Like Reflection, it pursues a sovereign-AI-friendly strategy, partners with NVIDIA (co-developed the first Nemotron base model), and competes directly for open-model developer mindshare and enterprise sovereign deployments.
• DeepSeek: DeepSeek is a Chinese AI lab whose open-weight models (DeepSeek-V3, R1) shocked the market in early 2025 with frontier-class performance at a fraction of training cost. It is the Pentagon's named "threat" that Reflection is positioned against — directly comparable as an open-weight frontier model lab competing for global developer adoption.
• Alibaba Qwen: Alibaba's Qwen model family is one of the most-downloaded open-weight LLM series on Hugging Face and a leading Chinese open-weight frontier. It competes head-to-head with Reflection for sovereign AI deployments in Asia, Africa, and emerging markets where Chinese models have established strong adoption.

Broad incumbents

• Meta AI (Llama): Meta's Llama family is the dominant open-weight foundation model line and the de facto benchmark for the open frontier. Llama's massive distribution via Meta's platforms, hyperscaler partnerships, and ecosystem of derivatives (Code Llama, Llama 2/3/4) make it Reflection's primary open-weight incumbent to displace for developer and enterprise mindshare.
• xAI: xAI develops Grok, a frontier LLM with partial open-weight releases (Grok-1), and operates one of the world's largest GPU clusters (Colossus/Memphis). xAI is a direct customer/competitor dynamic via the SpaceX Colossus 2 deal and competes for the same Pentagon IL6/IL7 slots as Reflection.
• OpenAI: OpenAI is the primary closed-lab incumbent Reflection's open-weight narrative is explicitly positioned against. Co-located on the Pentagon IL6/IL7 list, OpenAI defines the frontier capability bar and serves as the default enterprise LLM — meaning Reflection's open-weight thesis must overcome OpenAI's distribution, brand, and capability lead.
• Anthropic: Anthropic (Claude family) is a leading closed frontier lab and a direct competitor for enterprise LLM spend, government contracts, and frontier talent. It is named in Reflection's research page as one of the closed labs the open-weight strategy aims to provide an alternative to.

Emerging players

• Perplexity: Perplexity is an AI-native answer-engine startup that uses frontier open-weight models and is a fellow NVIDIA Nemotron Coalition member. It represents a downstream consumer of open-weight frontier models and a coalition partner for shared compute and research.
• Cohere: Cohere is an enterprise-focused foundation model company offering Command models and a platform for custom model deployment. It targets the same enterprise/government sovereign-AI buyers as Reflection with a comparable full-stack training and serving offering, though with a primarily proprietary (closed-weight) model approach.
• Thinking Machines Lab: Thinking Machines Lab is a frontier AI research startup founded by former OpenAI/CTO Mira Murati, a fellow Nemotron Coalition member. It competes with Reflection for elite talent, frontier model positioning, and large-scale NVIDIA compute access — making it a comparable emerging frontier lab rival.

Market position

Strengths5 records

Weaknesses4 records

Competitive moat6 records

Key risks6 records

Key highlights7 records

Customer concentration

• Xx.com/reflection_ai
• LinkedInlinkedin.com/company/reflection-ai

Revenue estimate

Valuation estimate

Number of profiles

Profiles2 records

Funding overview

Funding rounds4 records

Investors18 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 Reflection AI