Reflection AI

Built for customization and long-term control

Building super intelligent autonomous systems

Reflection raises $2B to build frontier open intelligence. Combining large-scale LLM training with advanced RL for accessible AI models.

The company’s co-founder Ioannis joined DeepMind as a founding engineer and helped create AlphaGo, the first system to surpass the human world champion in Go. The 2016 event was a turning point for the Reflection team’s view of superintelligence.

Reflection is a research and product company whose mission is to build open superintelligence. It is hiring across all roles.

The company will release open weights for its models.

Nvidia introduces the **OpenShell Reference Architecture**, a framework to secure and govern AI agents using its OpenShell runtime. The system enforces boundaries, monitors behavior, and integrates with Nvidia’s Vera CPUs and BlueField-4 DPUs for hardware-backed security. Early adopters include banks like BBVA and Santander, while SAP is integrating it with its Joule Studio runtime. OpenShell requires Nvidia hardware and Linux/Kubernetes infrastructure, with support from hyperscalers AWS, Google Cloud, and Microsoft Azure.

Oct 2026The RegisterNews article

Eric Schmidt holds roughly **1% of Anthropic**, a stake valued at tens of millions from the 2021 Series A round when the AI lab was worth **$461 million**. This compares to CEO Dario Amodei’s **2% ownership**, positioning Schmidt as a major shareholder ahead of Anthropic’s anticipated IPO. If the company reaches a projected **$2 trillion valuation**, Schmidt’s stake could add **$20 billion** to his net worth. He has previously invested in AI firms like Mistral and General Intuition through entities including Tomorrow Ventures.

Oct 2026TechCrunchNews article

Reflection AI releases Beam’s weights and post-training tools on Hugging Face, enabling developers to customize its 500-billion-parameter large language model. The company introduces quantization variants, training recipes for fine-tuning, and a drafter model for speculative decoding. Beam achieves near-top-tier performance with reduced compute and memory, supported by custom reliability optimizations and serving engine configurations.

Oct 2026SiliconANGLENews article

The drumbeat over how safe artificial intelligence is — or more to the point, isn’t — just keeps louder. In the wake of weeks of incidents involving rogue agents, with likely more to come, more government officials are calling for regulation, even if President Trump says it’s not needed. Several sta

Oct 2026SiliconANGLENews article