Local Chorus
Local news from local sources, read in your language.
Settled

Reflection AI releases first open-weight model Beam, taking on Chinese AI rivals

🇰🇷 South Korea 09:53 AI Business Official updated 18 h ago first reported by 연합뉴스

In short

Reflection AI, a US startup founded by two former Google DeepMind researchers, released its first open-weight model, Beam, designed for coding and AI agent tasks. The company said Beam performs comparably to Chinese rival Z.AI's GLM-5.2 on advanced reasoning benchmarks, beats Thinking Machines and Nvidia models, but trails some Chinese models. Reflection AI is backed by Nvidia and has tied up a sovereign AI partnership with Shinsegae Group in South Korea.

Read the full story 1 min read

Reflection AI, a US artificial intelligence startup founded by two former Google DeepMind researchers, has released its first open-weight model, called Beam, according to Yonhap and Chosun Ilbo. Open-weight models can be downloaded free of charge, and Yonhap said such models offer advantages in security and customization, plus more predictable pricing, compared with the closed models of OpenAI and Anthropic. [ 1 , 2 ]

Yonhap said Reflection describes Beam as designed to give enterprise customers and developers strong performance in coding and AI agent work, and as cheaper than many competing models. The company said Beam performs similarly to GLM-5.2, the flagship open-source model of China's Z.AI, on advanced reasoning benchmarks, beats Thinking Machines and Nvidia AI models, but lags somewhat behind Chinese models. [ 1 ]

Yonhap framed the launch as an addition to the line-up of US competitors trying to match free models released in China, and said Reflection aims to counter Chinese companies such as DeepSeek that have dominated the open-weight market. Bloomberg assessed that the release strengthens the US government's efforts to export American and allied AI abroad, Yonhap reported. [ 1 ]

Chosun Ilbo led its coverage with Nvidia's investment in Reflection AI and said the company claims the new model matches leading American and Chinese open-weight models. It dated the release to Oct. 5 local time and the company's founding to 2024. Nvidia's backing is also reported by Yonhap, which additionally cites support from other investors. [ 1 , 2 ]

Yonhap reported that earlier this year Reflection signed a strategic partnership agreement, or MOU, with South Korea's Shinsegae Group to build a sovereign AI factory. The company has also signed multi-billion-dollar contracts for the computing capacity needed to develop its models with Elon Musk's SpaceX and with the neocloud company Nebius Group, according to the same report. [ 1 ]

Why it matters

The release adds a US entrant to the open-weight market long dominated by Chinese firms such as DeepSeek, and Bloomberg assessed it as reinforcing Washington's push to export US and allied AI abroad. It also connects a US AI developer to Korean industry through a sovereign AI factory agreement with Shinsegae Group and to large computing contracts abroad.

Key facts

  • Reflection AI released its first open-weight model, named Beam, designed for coding and AI agent work. [ 1 , 2 ]
  • The company was founded by two former Google DeepMind researchers, and Chosun Ilbo dates the founding to 2024. [ 1 , 2 ]
  • Reflection said Beam performs similarly to Z.AI's open-source GLM-5.2 on advanced reasoning benchmarks and beats Thinking Machines and Nvidia AI models, while trailing some Chinese models. [ 1 ]
  • Reflection said Beam is cheaper than many competing models. [ 1 ]
  • Nvidia is an investor in Reflection AI, according to both outlets. [ 1 , 2 ]
  • Reflection signed a strategic partnership MOU with Shinsegae Group earlier this year to build a sovereign AI factory. [ 1 ]
  • Reflection signed multi-billion-dollar contracts with Elon Musk's SpaceX and neocloud company Nebius Group for computing capacity. [ 1 ]

Confirmed by several sources

  • Reflection AI released an open-weight AI model called Beam. [ 1 , 2 ]
  • The company was founded by two former Google DeepMind researchers. [ 1 , 2 ]
  • Nvidia has invested in Reflection AI. [ 1 , 2 ]

Still unclear

  • The exact date and time Beam was released. Chosun Ilbo says the model was released on Oct. 5 local time, while Yonhap reports it on Oct. 6 without giving a release date, and no precise time is stated.
  • Whether Beam's performance claims hold up independently. The benchmark comparisons are the company's own claims, reported by Yonhap and repeated in summary form by Chosun Ilbo, with no independent verification cited.
  • The year Reflection AI was founded. Only Chosun Ilbo states 2024; Yonhap does not give a founding year.

What local media are saying

Official sourcesYonhap, the state news agency, gave the fullest account, describing Beam's positioning against Chinese and US models, the open-weight market dominated by DeepSeek, the Shinsegae sovereign AI factory MOU, and the SpaceX and Nebius computing contracts, and citing Bloomberg's assessment of US export efforts. [ 1 ]
Business mediaChosun Ilbo led with Nvidia's investment in Reflection AI and framed the launch as a challenge to Chinese AI, repeating the company's claim of performance comparable to leading open-weight models. [ 2 ]

Timeline, local time

  1. Reflection AI releases Beam, dated by Chosun Ilbo to Oct. 5 local time; no time of day is given. [ 2 ]
  2. Yonhap publishes its report on the Beam release. [ 1 ]
  3. Chosun Ilbo publishes its report on the Beam release. [ 2 ]