Know
What changedGemini Decisions API scores up to 1,000 questions
Google introduced the Gemini Decisions API, powered by Gemini Decision 1, for closed-ended classification and scoring. A request can assess up to 1,000 independent questions against shared text, image, audio, video or PDF input; results include probabilities and confidence values rather than generated text.
Consider it for structured scoring or classification workloads where you need confidence values, not prose.
Qwen4-Max-Preview requires a Token Plan
Alibaba Cloud opened Qwen4-Max-Preview for text, image and video understanding, reasoning and tool calls. Access is limited to Token Plan subscribers—not pay-as-you-go users—and the preview is scheduled to end November 25, 2026.
Check your plan before testing; the preview has a stated end date and no pay-as-you-go access.
Databricks makes Genie Ontology generally available
Databricks made Genie Ontology, which represents relationships across business data, people and processes, generally available. Its announced second-generation Genie can run long tasks in the background, delegate to subagents and send alerts, but Always-on Genie, Genie Search and Ontology Manager are coming in preview or beta.
Ontology is available now; treat the always-on agent and related tools as forthcoming, not production-ready options.
Plurai simulates enterprise agent workflows
Plurai raised a $13 million seed round to build digital twins of company systems, including an agent’s accessible tools and knowledge. It maps possible workflows and uses simulations and synthetic data to evaluate edge cases; the company says it works alongside Claude Code, Codex, Copilot and Cursor.
If evaluating enterprise agents, compare simulated edge-case testing with your current workflow tests.
United plans to block agents from completing bookings
United says it will restrict AI bots from booking, ticketing or servicing flights, including through agencies and aggregators, and is revising its policies. Assistants may still show flight options, but customers must complete purchases themselves; the airline cites risks from account access and unintended transactions.
For travel agents, design for human checkout and avoid assuming United’s booking flow will permit bot completion.
Fractile raises $720 million for inference chips
British inference-chip startup Fractile raised a $720 million Series C, with the UK Sovereign AI Fund among the participants. The UK government put its valuation above £5 billion, while the FT and a Reuters source reported about $7 billion. Fractile pairs chips with a machine-learning compiler and plans first shipments next year.
Treat this as a future infrastructure option, not available capacity; no performance or pricing details were supplied.