Reflection AI open-sources 501B-parameter Beam model
Reflection AI has reportedly introduced Beam, an open-source large language model said to contain 501 billion parameters—potentially the largest such model made public. The announcement follows reports of a $6.3 billion agreement with SpaceX to rent Nvidia GB300 NVL72 appliances, which, if accurate, would suggest significant compute investments. The startup was previously valued at $25 billion in an undisclosed funding round.
The move appears to fit a growing pattern of startups open-sourcing AI models to boost adoption and acquisition prospects, as StartupReader observed earlier this month. If confirmed, Beam’s scale would surpass recent open-weight releases, though its practical impact remains to be seen. The company’s primary product—a mental health-focused AI journal—seems distant from this infrastructure effort, leaving its broader intentions ambiguous.
Should the SpaceX deal hold, it could place Reflection AI among the most prominent private buyers of Nvidia’s GB300 clusters. For now, it’s unclear whether Beam marks a one-time experiment or the start of a larger open-source push. Observers will be watching for further developments.
Sources: siliconangle.com
“Reflection AI’s reported release of a 501B-parameter open-source model may signal a new scale in open-weight AI, though its strategy remains unclear amid compute deals and a lofty valuation.”
Read the original reporting
The outlets below did the original reporting.
- Reflection AI debuts open-source Beam model with 501B parameters — siliconangle.com
Related briefs
- Pluto Security raises $42M as AI agent security funding surge continues
- TypeSafe AI raises $870M in largest US startup round of the week
- Clueso raises $10 Mn Series A for AI-powered enterprise video agents
- Mecka’s $60M round fuels robotics training data race
- NuSynth raises ₹36.5 Cr to accelerate AI-driven chemical R&D
This brief was drafted automatically from the sources above and published under our editorial policy. Spotted an error? Tell us.