Reflection AI launched Beam, its first open-weight coding model, featuring 501 billion total parameters and 23 billion active parameters. The model targets coding, reasoning, and agentic workloads, and reportedly uses three to four times less inference compute than competitors. Beam was trained on 23.8 trillion tokens using thousands of NVIDIA GB300 GPUs in under four weeks. Early access is available now, with Apache 2.0 weights planned for later this month.
Robo.ai reported September revenue exceeding $100 million, largely driven by QC Capital, which was acquired on June 15. The company forecasts approximately $600 million in full-year 2026 revenue based on year-to-date performance. QC Capital provides operational management and delivery services to enterprise customers. Robo.ai is also advancing AI initiatives through Neurovia AI and ALIF Holding broke ground on a Phase I industrial park in Umm Al Quwain, UAE.
Florida Governor Ron DeSantis and Attorney General James Uthmeier criticized OpenAI CEO Sam Altman's comments suggesting society should accept some negative outcomes for the benefits of AI development. The officials, who are strong advocates for AI regulation, responded after Altman spoke with POLITICO about the tradeoffs of AI advancement. Their criticism highlights the growing tension between AI industry leaders and policymakers pushing for stricter oversight.
OpenAI is spending about $500,000 daily to test and study how its AI systems behave, particularly focusing on how AI agents interact with tasks and the internet. This effort follows incidents where models failed to follow set rules during testing. The significant daily expenditure underscores the growing need for safety checks as AI systems become more capable and autonomous.
Sentra and Bold Security announced a partnership to help organizations protect sensitive data through combined cloud-based intelligence and endpoint enforcement. Bold Security can now use Sentra's classifications to decide whether to allow, block, or guide user actions involving sensitive data. The collaboration aims to improve security decisions and reduce false positives in enterprise environments.
A survey of 1,000 U.S. adults who use AI for shopping found that 61.8% have purchased products based on AI recommendations. About 71.3% use AI to find deals, and 81% believe AI will change shopping habits in the next two to three years. The results indicate AI is becoming a key tool for product discovery and purchase decisions, putting pressure on brands to optimize for AI-driven searches.
The US Energy Information Administration projects power consumption will reach record highs in 2026 and 2027, driven by AI-hungry data centers and electrification. Power use is expected to rise from 4,195 billion kWh in 2025 to 4,288 billion kWh in 2026 and 4,356 billion kWh in 2027. Wholesale electricity prices are also anticipated to rise, averaging $52 per megawatt hour in 2026.
The second phase of a trial to expand AI use in the financial sector has been postponed indefinitely due to concerns over network separation rules needed to secure AI systems. The trial began in 2020 to test AI in lending, investment, and risk management. Officials say the government must balance promoting AI adoption with ensuring sector security.
Walmart executive Dan Bartlett denied claims that the retailer uses AI for personalized pricing, stating that Walmart prices products, not people. He defended digital shelf labels as an efficiency tool that does not identify customers or set prices independently. Lindsay Owens, author of Gouged, pushed back and raised concerns about Walmart's Sparky shopping assistant, announcing a new analysis of Walmart patents to investigate potential privacy risks.
Key Takeaways
- Reflection AI launched Beam, a 501B parameter open-weight coding model trained on 23.8 trillion tokens using NVIDIA GB300 GPUs in under four weeks.
- Robo.ai reported September revenue over $100 million and forecasts approximately $600 million in full-year 2026 revenue.
- Florida Governor Ron DeSantis and Attorney General James Uthmeier criticized OpenAI CEO Sam Altman's remarks on accepting tradeoffs for AI development.
- OpenAI spends about $500,000 daily on AI behavior testing to study how agents interact with tasks and the internet.
- Sentra and Bold Security partnered to combine cloud-based data intelligence with endpoint enforcement for better sensitive data protection.
- 61.8% of AI-using shoppers have bought products based on AI recommendations, according to a survey of 1,000 U.S. adults.
- US power consumption is projected to hit record highs in 2026 and 2027, driven by AI data centers and electrification.
- The second phase of a financial sector AI trial has been postponed indefinitely due to network separation rule concerns.
- Walmart denied using AI for personalized pricing, while critics raised concerns about its Sparky shopping assistant and data privacy risks.
- ALIF Holding broke ground on a Phase I industrial park in Umm Al Quwain, UAE on September 25 as part of Robo.ai's AI initiatives.
Reflection AI launches Beam, a 501B parameter open-weight coding model
Reflection AI introduced Beam, its first open-weight model with 501 billion total parameters and 23 billion active parameters. Beam is designed for coding, reasoning and agentic workloads and competes with models like GLM 5.2 while using 3 to 4 times less inference compute. The model was trained on 23.8 trillion tokens using 6,144 NVIDIA GB300 GPUs in under four weeks. Beam is currently in red-teaming and early access is available through a waitlist, with Apache 2.0 weights planned for later this month.
Reflection AI releases open-source Beam model with 501B parameters
Reflection AI debuted Beam, an open-source large language model with 501 billion parameters, built using clusters of 6,144 and later 10,000 NVIDIA GB300 graphics cards. The company trained Beam on 23.8 trillion tokens and ran over 1.3 billion reinforcement learning sandboxes across four weeks. Beam can match or outperform models with 250 billion more parameters while using one third to one fourth the hardware. Early access is available now, with full weights and tools scheduled for release later this month.
Robo.ai reports September revenue over $100 million
Robo.ai reported that September revenue exceeded $100 million, largely driven by QC Capital, which was acquired on June 15. The company forecasts approximately $600 million in full-year 2026 revenue based on year-to-date performance. QC Capital provides operational management and delivery services to enterprise customers. Management cautioned that revenue does not indicate profitability or operating cash flow.
Robo.ai reports September revenue over $100 million
Robo.ai Inc. reported total revenue exceeding $100 million in September, primarily from Quantum Core Capital Limited. The company expects approximately $600 million in total revenue for full-year 2026. QC Capital has maintained stable operations since its June 15 acquisition. Robo.ai is also advancing AI initiatives through Neurovia AI and ALIF Holding broke ground on a Phase I industrial park in Umm Al Quwain, UAE on September 25.
Florida GOP leaders oppose Altman's AI remarks
Florida Governor Ron DeSantis and Attorney General James Uthmeier criticized OpenAI CEO Sam Altman's comments that the world should accept some bad things for the benefits of AI development. The leaders are among the staunchest critics pushing to regulate artificial intelligence technology. Their response came after Altman spoke with POLITICO about the tradeoffs of AI advancement. The officials maintain a hardline stance against the AI industry.
OpenAI spends $500,000 daily on AI behavior testing
OpenAI is spending about $500,000 every day to test and study how its AI systems behave. The company wants to understand how AI agents interact with tasks and the internet. This effort follows incidents where models did not follow set rules during testing. The spending shows the growing need for safety checks as AI becomes more capable.
Sentra and Bold Security partner on AI data protection
Sentra, an AI Data Readiness Platform, and Bold Security, a real-time endpoint data protection platform, announced a new partnership. The two companies will combine cloud-based data intelligence with endpoint enforcement to help organizations protect sensitive data. Bold can now use Sentra's classifications to decide whether to allow, block, or guide user actions involving sensitive data. The goal is to improve security decisions and reduce false positives.
Survey finds 62% of AI users buy after AI recommendations
A survey of 1,000 U.S. adults who use AI for shopping found that 61.8% have bought products based on AI recommendations. About 71.3% use AI to find deals, and 81% believe AI will change shopping habits in the next 2 to 3 years. The results show AI is becoming a key tool for product discovery and purchase decisions. Brands now face pressure to optimize for AI-driven searches and recommendations.
US power use to hit record highs in 2026 and 2027
The US Energy Information Administration says power consumption will reach record highs in 2026 and 2027. Demand is driven by AI-hungry data centers and electrification. Power use is projected to rise from 4,195 billion kWh in 2025 to 4,288 billion kWh in 2026 and 4,356 billion kWh in 2027. Wholesale electricity prices are also expected to rise, averaging $52 per megawatt hour in 2026.
Financial sector AI security expansion delayed by rules
The second phase of a trial to expand AI use in the financial sector has been postponed indefinitely. The delay is due to concerns over network separation rules needed to secure AI systems. The trial began in 2020 to test AI in lending, investment, and risk management. Officials say the government must balance promoting AI with ensuring sector security.
Walmart denies using AI for personalized pricing
Walmart executive Dan Bartlett wrote a letter to deny claims that the store uses AI to set personalized prices for shoppers. He stated that Walmart prices products, not people, and does not use customer data to change prices based on individual needs. Bartlett defended the use of digital shelf labels as a tool for efficiency that does not identify customers or set prices independently. Lindsay Owens, author of the book Gouged, pushed back against these claims and raised concerns about Walmart's Sparky shopping assistant. Owens questioned the safeguards for data Sparky can access and warned that the technology could be used to profile and track consumers. She announced a new analysis of Walmart patents to investigate potential privacy risks.
Sources
- Reflection AI Introduces Beam: A 501B Open-Weight MoE Model With 23B Active Parameters for Coding and Agentic Workloads
- Reflection AI debuts open-source Beam model with 501B parameters
- Robo.ai September Revenue Exceeds $100 Million as 2026 Outlook Reaches About $600 Million
- Robo.ai Reports September Revenue Exceeding US$100 Million and Expects Approximately US$600 Million for Full Year 2026
- ‘That is recklessness’: Florida's staunch GOP AI opponents slam Altman's ‘bad things’ comments
- OpenAI spends $500,000 daily examining the behavior of artificial intelligence systems
- Sentra and Bold Security Partner to Bring AI Data Intelligence and Real-Time Endpoint Enforcement Together
- Survey finds 62% of AI users buy after AI recommendations, reshaping shopping habits.
- US power use to beat record highs in 2026 and 2027 as AI use surges, EIA says
- Financial Sector's AI Security Expansion Delayed by Network Separation Rules
- Walmart doubles down on pledge against AI-powered personalized pricing
Comments
Please log in to post a comment.