Jump to content

Reflection AI

Edit links
From Wikipedia, the free encyclopedia

Reflection AI
TypePrivately held company
IndustryArtificial intelligence
Founded2024; 2 years ago (2024)
FoundersMisha Laskin
Ioannis Antonoglou
HeadquartersNew York City
ProductsBeam,[1] Asimov[2]
Number of employees
320 (2026) Edit this on Wikidata
Websitereflection.ai

Reflection AI is an American artificial intelligence company that develops open foundation models and software agents for AI-assisted software development. Founded in 2024 by former Google DeepMind researchers Misha Laskin and Ioannis Antonoglou, the company initially focused on tools that automate software development.[3] As of June 2026, Reflection's valuation is $25 billion.[4]

Reflection positions itself as a US open weights AI lab, as an alternative to both Chinese labs and the proprietary models of leading US labs.[5] Reflection reportedly plans to use revenue sharing licenses with its large language models (LLMs).[6] In October, Reflection announced its first LLM, Beam, with 501 billion parameters.

Reflection has compute contracts with SpaceXAI, Nebius Group, and Shinsegae, and under the US government's Genesis Mission is set to provide models for the US national laboratories. Reflection is headquartered in New York City, with offices in San Francisco and London.

Technology

[edit]

Reflection's work combines large language model training, reinforcement learning, agentic AI, and software engineering automation. The company says it is developing open foundation models and advancing both pre-training and post-training systems, with an emphasis on reinforcement learning at scale.[7]

In October 2025, Reflection said it had built a large-scale large language model and reinforcement learning platform capable of training mixture of experts models at frontier scale, first applying the approach to autonomous coding and later expanding it toward general agentic reasoning.[8]

Reflection's first product was Asimov, a code-comprehension agent for engineering teams. Wired reported in July 2025 that Asimov reads source code, emails, Slack messages, project updates and documentation to answer questions about how software systems are built.[9]

On 5 October 2026 Reflection announced its first model, Beam, with 501 billion parameters, 23 billion active. It was positioned as a “workhorse” that performs better than other Western open models at a much lower cost of inference compute. Beam performed similarly in benchmarks to GLM 5.2, a previous-generation open model by Chinese firm Z.ai. As of 10 October, the model remains in early access.[10][11][5]

Funding and partnerships

[edit]

In March 2025, Reflection emerged from stealth with $130 million in financing, including a $25 million seed round and a $105 million Series A round. The financing valued the company at about $545 million.[12]

In October 2025, Reflection raised $2 billion in a funding round that valued the company at $8 billion. Investors in the round included Nvidia, Eric Schmidt, Citigroup, 1789 Capital, Lightspeed Venture Partners and Sequoia Capital.[13][14]

In March 2026, Reflection and Shinsegae signed a memorandum of understanding to build one of the country's largest facilities powering AI models, a 250-megawatt data center in South Korea, with Nvidia supplying tens of thousands of chips.[15][16][17]

In May 2026, Axios reported that Reflection was partnering with the United States Department of Energy to support the Genesis Mission, a federal scientific research initiative, and would serve as the AI model provider for the U.S. National Laboratories.[18]

In June 2026, Axios reported that Reflection had signed a compute agreement with SpaceXAI for access to chips and hardware from the SpaceX Colossus data center. Under the agreement, Reflection would pay $150 million per month starting July 1, 2026, through 2029, after an initial ramp period.[19]

Following talks with investor JPMorgan Chase, Reflection was valued at $25 billion as of June 2026.[20]

In July 2026, Reflection and Nebius signed an agreement for over $1 billion in computing power, allowing Reflection access to Nvidia's GB300 chips through 2029.[21][22]

As of October 2026, Reflection lists its headquarters in New York City, with other offices in San Francisco and London.[23]

See also

[edit]

References

[edit]
  1. ↑ Nazzarro, Miranda (October 5, 2026). "Reflection AI releases first open model to rival China". The Hill. Retrieved October 12, 2026.
  2. ↑ Zhan, Stephanie; Curnin, Charlie (July 16, 2025). "Reflection AI Launches Asimov: Breakthrough Agent for Code Comprehension". Sequoia Capital. Retrieved June 22, 2026.
  3. ↑ "Nvidia-backed Reflection AI raises $2 billion in funding, boosts valuation to $8 billion". Reuters. October 9, 2025. Retrieved June 22, 2026.
  4. ↑ Bosa, Deirdre (June 22, 2026). "SpaceX signs compute deal with open-source AI startup Reflection". CNBC.
  5. 1 2 Forlini, Emily. "Reflection AI unveils its first open model, Beam. Could it be America's best chance to compete with China?". Fortune. Retrieved October 10, 2026.
  6. ↑ Bellan, Rebecca (October 9, 2025). "Reflection AI raises $2B to be America's open frontier AI lab, challenging DeepSeek". TechCrunch. Retrieved June 22, 2026.
  7. ↑ "Research". Reflection AI. Retrieved June 22, 2026.
  8. ↑ "Building Frontier Open Intelligence Accessible to All". Reflection AI. October 9, 2025. Retrieved June 22, 2026.
  9. ↑ Knight, Will (July 16, 2025). "Former Top Google Researchers Have Made a New Kind of AI Agent". Wired. Retrieved June 22, 2026.
  10. ↑ Gold, Ashley (October 5, 2026). "Reflection AI raises $2B to be America's open frontier AI lab, challenging DeepSeek". Semafor. Retrieved October 6, 2026.
  11. ↑ Bellan, Rebecca (October 5, 2026). "Reflection debuts Beam, an open-weight AI model to rival Chinese models at lower compute cost". TechCrunch. Retrieved October 6, 2026.
  12. ↑ Wiggers, Kyle (March 7, 2025). "Ex-DeepMind researchers' new startup aims for superintelligence". Bloomberg News. Retrieved June 22, 2026.
  13. ↑ "Nvidia-backed Reflection AI raises $2 billion, boosts valuation to $8 billion". Reuters. October 9, 2025. Retrieved June 22, 2026.
  14. ↑ Clark, Kate; Schechner, Sam (August 2, 2026). "The Race to Build an American Alternative to Cheap AI From China". The Wall Street Journal.
  15. ↑ Ramkumar, Amrith (March 17, 2026). "Nvidia-backed AI startup to spend billions on Korea data center to combat China". The Wall Street Journal.
  16. ↑ Park, Jae-hyuk (March 17, 2026). "Shinsegae teams up with US tech firm to build Korea's largest AI data center". The Korea Times.
  17. ↑ Gooding, Matthew (March 19, 2026). "Shinsegae Group and Reflection AI to build 250MW sovereign AI data center in South Korea". DatacenterDynamics.
  18. ↑ Gold, Ashley (May 22, 2026). "Exclusive: Reflection AI to power Genesis Mission". Axios. Retrieved June 22, 2026.
  19. ↑ Mills, Madison (June 22, 2026). "Open-source AI gets more compute from SpaceX". Axios. Retrieved June 22, 2026.
  20. ↑ Jin, Berber; Clark, Kate (March 25, 2026). "Nvidia-backed startup seeking to counter Chinese AI eyes $25 billion valuation". The Wall Street Journal.
  21. ↑ Soni, Aditya (July 14, 2026). "AI startup Reflection signs over $1 billion computing deal with Nebius". Reuters.
  22. ↑ Bass, Dina (July 14, 2026). "Nebius to Sell $1 Billion in AI Capacity to Startup Reflection". Bloomberg.
  23. ↑ "Reflection: About | LinkedIn". LinkedIn. Retrieved October 10, 2026.
[edit]
Retrieved from "https://en.wikipedia.org/w/index.php?title=Reflection_AI&oldid=1379939634"