Reflection AI company profile: valuation, funding & revenue

This Reflection AI company profile covers the company's valuation, revenue, funding rounds, investors and history. Reflection AI works in artificial intelligence, with a focus on open frontier models and coding agents. It was founded in 2024 and is headquartered in New York, New York, United States. Its latest valuation on record is US$8 billion (October 2025). Every figure links to a public source, and anything the company hasn't disclosed is marked as such.

Revenue

Funding history

Reflection AI timeline

Key dates from the sources on this page, oldest first. Metric snapshots are listed after the funding events.

• 2024 — Reflection AI is founded in New York by Misha Laskin (CEO) and Ioannis Antonoglou.
• Mar 2025 — Series A: US$130M, valuing the company at About US$545M. Investors: Lightspeed, CRV.
• Oct 2025 — Series B: US$2B, valuing the company at US$8B. Investors: Nvidia (lead); Eric Schmidt, Citi, 1789 Capital, Lightspeed, Sequoia.
• 2026 — Reported talks: Reported to be raising about US$2.5 billion at a valuation of about US$25 billion; not confirmed as closed.

Investors

Overview

Reflection AI is building frontier-scale AI models that it intends to release with open weights, positioning itself as an American answer to open model developers such as DeepSeek, Alibaba's Qwen team and Moonshot AI in China, and Mistral AI in Europe. The company began by building autonomous coding agents, software that can understand large codebases and carry out programming tasks, and then broadened its mission to training general-purpose frontier models that companies, researchers and governments can download, inspect, customise and run on their own infrastructure.

Misha Laskin, the CEO, and Ioannis Antonoglou founded Reflection AI in 2024. Both were senior researchers at Google DeepMind. Antonoglou was a key contributor to AlphaGo, AlphaZero and MuZero, the reinforcement learning systems that mastered Go, chess and other games, and later worked on Gemini. Laskin worked on reinforcement learning and on training methods for Gemini. Their combined experience in reinforcement learning, the technique of training models through trial and reward, is central to Reflection's approach to building models that can reason and act, not only generate text.

From coding agents to open frontier models

Reflection's first product, Asimov, was a code comprehension agent designed to help engineering teams understand large, complex codebases by reading code, documentation and discussions and answering questions about how systems work. The founders saw coding as the best proving ground for autonomous agents, because software tasks can be tested automatically and because companies spend enormous sums on engineering. In 2025, Reflection announced a larger ambition: training frontier-scale models and releasing them openly. It argued that the most capable open models were increasingly coming from China, and that the United States needed a strong open model developer so that businesses and governments could build on technology they could fully control.

Why open-weight models matter

Closed models, such as those from OpenAI and Anthropic, are accessed only through their makers' services. Open-weight models can be downloaded and run anywhere: inside a company's own data centre, in a specific country for data residency, or on specialised hardware. That gives users more control over cost, privacy and customisation, and lets researchers study how models work. Open models from Meta, Mistral, DeepSeek and Qwen have become widely used, and Chinese labs in particular have released highly capable open models. Policymakers and industry groups in the US have raised concerns about American companies and agencies depending on foreign open models, which created an opening that Reflection is targeting.

Funding

Reflection raised about US$130 million in early 2025 at a valuation of roughly US$545 million, according to PitchBook data cited in reports, with backing from Lightspeed and CRV. In October 2025 it raised US$2 billion at an US$8 billion valuation, a jump of about fifteen times in seven months. Nvidia led the round, with participation from former Google CEO Eric Schmidt, Citi, 1789 Capital, Lightspeed and Sequoia Capital. In 2026, Reuters reported that Reflection was expected to raise about US$2.5 billion at a valuation of about US$25 billion. We haven't seen that round confirmed, so the data above uses the October 2025 figure and lists the reported talks separately.

Business model

Open-weight model companies face a central question: how to make money if the core product is given away. Common approaches include selling access to hosted versions of the models through an API, offering enterprise support, fine-tuning and deployment services, licensing larger or specialised models on commercial terms, and partnering with governments and cloud providers on sovereign AI projects. Reflection hasn't disclosed revenue, so we show it as not disclosed. Its value today rests mainly on its team, its compute resources and the strategic importance of an American open model developer.

Competition

Reflection competes with open model developers including Meta's Llama team, Mistral AI, DeepSeek, Alibaba's Qwen, Moonshot AI and others, and indirectly with closed frontier labs such as OpenAI, Anthropic and Google DeepMind. In coding agents, it competes with Cognition, Cursor, Claude Code, Codex and GitHub Copilot. Its advantages are its founders' reinforcement learning expertise, very large funding and Nvidia's strategic backing.

Reflection AI at a glance

Reflection AI is a New York AI lab founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou. It started with coding agents and is training open-weight frontier models. It raised US$2 billion at an US$8 billion valuation in October 2025, led by Nvidia, and was reported in 2026 to be raising at about US$25 billion.

Timeline

• 2024: founded by Misha Laskin and Ioannis Antonoglou after leaving Google DeepMind.
• March 2025: emerges publicly with about US$130 million in funding.
• July 2025: launches the Asimov code comprehension agent.
• October 2025: raises US$2 billion at US$8 billion, led by Nvidia, and announces its open frontier model mission.
• 2026: reported to be raising about US$2.5 billion at about US$25 billion.

Why Nvidia invests in AI labs

Nvidia has become one of the most active investors in AI startups, backing model developers, cloud providers and application companies. Its investments help ensure strong demand for its chips, since labs spend much of the money they raise on computing. Strong open models also expand the overall market for AI infrastructure, because companies running open models need GPUs, either their own or through cloud providers. For Reflection, Nvidia's backing brings capital, credibility and closer access to the latest hardware.

The cost of training frontier models

Training a frontier model requires tens of thousands of advanced GPUs running for months, along with large teams of researchers and engineers and huge volumes of data. Costs for the largest training runs have reached hundreds of millions of dollars and continue to rise. Reflection's US$2 billion round gives it the resources to compete at this scale, but frontier labs typically need to raise again quickly as model sizes and compute requirements grow. That's why its reported 2026 talks at a much higher valuation matter for its long-term plans.

Risks

• Execution: it must deliver open models that match or beat leading Chinese and closed models.
• Monetisation: earning substantial revenue from open models is harder than from closed APIs.
• Capital needs: frontier training requires repeated large funding rounds.
• Policy: rules on releasing powerful open models could change.

Questions for enterprises evaluating open models

• How does the model perform on your tasks compared with closed alternatives?
• What licence terms apply to commercial use and modification?
• What infrastructure is needed to run it efficiently?
• What support and security updates will the developer provide?

Latest developments (October 2026)

Reflection AI, which is building open frontier models with backing from Nvidia, was reported in 2026 to be raising about US$2.5 billion at roughly US$25 billion before the new money, according to Shopifreaks, citing earlier Wall Street Journal reporting. We have not seen confirmation that the round closed, so our profile keeps the October 2025 figure of US$8 billion. The company is pitching its models to governments seeking sovereign AI, but had not released a public frontier model by mid-2026. See our new AI labs guide.

What to watch

Watch for Reflection's first open frontier model release and its benchmark results, confirmation of new funding, partnerships with governments and cloud providers, and how US policy on open models develops. Our guide to the most valuable AI startups compares Reflection with other labs.

Takeaways

Enterprises may gain a strong American open-weight option for private deployment. Founders can see how a clear strategic narrative attracts very large rounds. Investors should weigh a fast-rising valuation against unproven models and revenue.

Sources

Every figure on this page comes from the public sources below. Where sources disagreed, we used the company's own announcement or a primary news report. Private-company data changes; check the date beside each figure.

1. AI Business
2. Yahoo Finance (citing Reuters)