Capital, compute economics, and evaluation moved to the center of the AI market. Mistral announced a €3 billion Series D led by Samsung, while new public TPU benchmarking challenged NVIDIA's cost-per-token advantage and Jensen Huang framed accelerated compute as a revenue-producing asset.
Agent capability is being measured more rigorously. Artificial Analysis added Terminal-Bench 4.0 and a private workflow-automation benchmark, while practitioners highlighted stronger scientific review, interactive visualization, and video generation.
Safety and governance remain geopolitical issues: the United States and China are preparing for bilateral discussions focused on AI-enabled cyberattacks and model distillation. Researchers are also probing whether language models use confidence to guide behavior, not merely report it.
Mistral said the financing is the largest equity round ever raised by a European technology company and values it above €21 billion. The company plans to expand its infrastructure, products, and deployment capacity, reinforcing Europe's effort to sustain an independent frontier-model supplier.
Officials are planning talks on AI-directed cyberattacks and model distillation ahead of a broader leaders' summit. The agenda could create a narrow channel for cooperation between the two largest AI powers even as their competition over chips, models, and national security intensifies.
Intelligence Index v4.3 replaces an older terminal benchmark with Terminal-Bench 4.0 and adds AutomationBench-AA, a held-out workflow benchmark. Private test sets and more realistic business tasks matter because public benchmarks are increasingly vulnerable to contamination and optimization that does not translate into operational performance.
SemiAnalysis' Dylan Patel announced an open, continuously running benchmark of Google's TPUs across multiple models and scenarios. The initial claim that TPUs deliver better cost per token than NVIDIA's B200 and B300 adds useful public evidence to a market where accelerator economics are often opaque.
MiniMax highlighted an open-source acceleration effort from NVIDIA's Sol team, with access through the Reactor API before users commit GPU capacity. Faster-than-playback generation could materially improve interactive video workflows and lower the infrastructure barrier for teams evaluating generative-video systems.
Selected posts from the edition, with links to the originals.
“Today marks a major step for Mistral: we're announcing a €3B Series D, the largest equity round ever raised by a European tech company, just three years after launch.”
View post on X ↗“NVIDIA compute is fungible, durable and highly rentable. It is a productive, revenue-generating asset.”
View post on X ↗“New video generation acceleration for MiniMax-H3 from the NVIDIA Sol team! Faster than playback, fully open-sourced, and available via the Reactor API.”
View post on X ↗“Astra for visualizing how a drug works.”
View post on X ↗“Announcing Artificial Analysis Intelligence Index v4.3, upgrading Terminal-Bench to 4.0 and adding AutomationBench-AA, an agentic workflow automation benchmark with a private test set.”
View post on X ↗“I would not ‘declare AGI’ until we have AI that is, at last, capable of invention — conceptual breakthroughs, novel insights.”
View post on X ↗“We are excited to bring the first open benchmarking of Google's TPUs to the world… $/token is better than B200 and B300.”
View post on X ↗“If you are an academic, you should try throwing a paper into GPT-6 Pro and asking ‘review this for me.’ It is now great at finding real issues while balancing those with a paper's strengths.”
View post on X ↗“Memory is one of the most challenging things to get right when building agents… This paper focuses on whether memory is portable.”
View post on X ↗“Do LLMs passively report confidence—or do they use it to guide behaviour? Our Nature Machine Intelligence paper provides causal evidence.”
View post on X ↗