This brief covers AI news from 2026-09-10 UTC.
Open-weight frontier models land from Alibaba and Ant Group, Google Cloud puts agents inside the editor, and two agent startups post nine-figure rounds.
Alibaba opens Qwen3.8 with 2.4T parameters and 95B active
What happened: Alibaba’s Qwen team released Qwen3.8-2.4T-A95B, the first open-weights release of a Qwen-Max-class model. It uses a hybrid attention architecture with 2.4 trillion total parameters and 95 billion activated per token, and was demonstrated on Amazon SageMaker HyperPod with vLLM on NVIDIA B300 GPUs.
Why it matters: A frontier-class model you can self-host changes the build-versus-rent math for teams running heavy agentic and reasoning workloads.
Ant open-sources Ling-3.0-flash-Fin for finance workflows
What happened: Ant Group announced the open-sourcing of Ling-3.0-flash-Fin at the 2026 Inclusion Conference on the Bund. The Mixture-of-Experts model has 124 billion total parameters and activates 5.1 billion per token, was co-developed with financial institutions, and targets investment research and financial analysis.
Why it matters: A 5.1B-active model tuned for compliance-heavy tasks gives fintech builders a cheap option for retrieval and analysis pipelines that need traceable sources.
Google Cloud previews Data Agent Kit for editors and CLIs
What happened: Google Cloud introduced Data Agent Kit, a preview product combining Model Context Protocol servers with agent skills so agents can query data systems and return results inside code editors and command lines. It ships as an extension for VS Code forks including Antigravity IDE and Cursor, and as a plugin for Antigravity 2.0, Antigravity CLI, Claude Code and Codex.
Why it matters: Data work moves into the IDE, so developers can answer cross-system questions without hopping between BigQuery, Postgres and storage consoles.
Clay raises $115M at $7.1B valuation for sales agents
What happened: Clay said it raised $115 million at a $7.1 billion valuation, more than twice its valuation a year earlier. The New York company builds AI agents that automate sales and marketing tasks, and counts Google and Anthropic among its customers.
Why it matters: Another nine-figure round into agent tooling shows investors still funding go-to-market automation, a segment where builders compete on integrations and data access.
Harvey hits $15.6B valuation with $550M round
What happened: Legal AI startup Harvey reached a $15.6 billion valuation with a $550 million funding round, according to Bloomberg. The company was co-founded by Winston Weinberg and Gabe Pereyra.
Why it matters: Legal AI keeps pulling large checks, a signal that vertical agents with proprietary domain workflows can command premium valuations.
Meta acquires Swedish startup Stilla.ai for Business Agent
What happened: Meta has acquired Swedish AI startup Stilla.ai to accelerate development of Meta Business Agent, which facilitates transactions on WhatsApp, Messenger and Instagram, Axios reported. Stilla was founded in Stockholm in 2024 and raised $5 million in pre-seed funding when it emerged from stealth in January.
Why it matters: Meta is buying commerce capability for the messaging agents that more than a million businesses already use to handle customer conversations.
Google puts $15B into Finnish AI infrastructure and nuclear power
What happened: Google said it will invest $15 billion in AI infrastructure in Finland and buy nuclear power there, Reuters reported. The announcement covers data center buildout and a nuclear power purchase in the country.
Why it matters: Power procurement is now part of the AI buildout, and where capacity lands shapes latency and cost for European workloads.