Google has a new frontier model, but most customers will wait while trusted testers find out where its extra power helps and hurts.
Today’s issue covers Gemini 4 Argon, finance’s expanding AI role, and the wide gap between spending and measured results.
Google announced Gemini 4 Argon yesterday, then put a waiting room in front of it. Trusted cyber defenders get first access while Google tests safeguards before a wider release.
The introductory price is $2 per million input tokens and $10 per million output tokens. Cached input costs 95% less, and the model can produce up to one million tokens in a single job.
Google says its own teams are already using Argon on long coding and research tasks. The early results are unusually concrete:
Agents found memory savings that freed more than 300 TiB across Google’s data centres.
One video-decoding rewrite ran 2.7 times faster than the earlier Rust version.
Wiz used Argon to find a critical flaw in hospital software that previous models had missed.
The model also topped Google’s tests for multi-step business work, including finance, legal research and tax. Those results come from Google and partners, so independent users still need to test them.
The release plan says as much as the benchmarks. Google thinks Argon can handle longer, more consequential jobs, but it is widening access in stages rather than opening the doors on day one.
Founders context: Do not plan around a model your team cannot access yet. When Argon opens more broadly, test one costly research, coding or document-heavy job against your current model and compare the full bill, review time included.
AI spending is moving from the innovation budget to the finance desk. In a new IBM and Oxford Economics survey, 62% of 1,500 CFOs said their role now includes more technology or AI strategy.
The extra authority has arrived before the systems are ready. Only 6% said finance has embedded AI across its work and decisions at scale.
The measurement gap is just as sharp:
48% said finance tracks value created by AI and moves capital accordingly.
8% said finance sets company-wide AI value goals with automatic funding triggers.
6% allow AI to recommend or execute budget reallocations within set limits.
IBM also identified a smaller group of “AI-first” finance teams. Their organisations reported 23% higher revenue growth from 2022 to 2024 and approved new AI funding 15% faster than peers.
The survey shows an association. It does not prove that giving finance more control causes growth.
Companies are still asking the people who guard the budget to decide which AI projects deserve more of it.
Founders context: Put a finance owner beside every material AI project. Agree the cost, the business measure and the stop date before work begins, then move money toward the few uses that beat the old process.
The AI boom is filling data centres faster than companies are filling spreadsheets with results. Andreessen Horowitz’s latest market review finds demand for computing power still outrunning supply, even for older chips.
Technology contributed about 76% of S&P 500 earnings growth in 2026 through late August, according to the review.
Big technology companies are turning cash into chips, power and networking, with debt taking a larger share of the bill.
Adoption inside ordinary companies looks much thinner:
Nearly 30% of S&P 500 companies report some quantifiable AI effect.
About 2% disclose a tracked metric for that effect.
Only a small share of agent users deploy them at meaningful scale.
Roughly 2% of US households paid for an AI service as of April.
The figures leave two stories running at once. Infrastructure suppliers are serving a market with no shortage of money, while most buyers are still working out which uses deserve a permanent line in the budget.
That gap is a business opportunity. Products that cut through company systems, finish a whole job and show the saving have more room than another assistant that produces activity without a number attached.
📡 Further Context
🧬 Google watermarks AI-designed proteins. SynthID Bio adds a detectable signature while preserving a protein’s biological function, giving researchers a way to trace synthetic designs.
🤒 Google tops a CDC flu-forecasting evaluation. Its model ranked first for predicting flu-related hospital admissions during the 2025-26 season.
🏛 The White House renames AI. An executive order tells federal agencies to use “Super Intelligence” in official documents, changing the label without changing the technology.
🔗 Bain finds a $100 billion cross-system software gap. Its thesis targets work that still needs people to copy information and decisions between separate tools.
🟣 OpenAI introduces always-on Dots. The DevDay product turns GPT-6 Astra into agents that can monitor work and continue tasks between sessions.
🤖 AI Tool Context
Gemini Skills: Reusable instructions for repeated work inside Gemini, replacing Gems.
OpenAI Dots: Always-on agents that can monitor and carry work between sessions.
GPT-6.1 Sol: A lower-cost model aimed at coding, computer use and professional work.
Gemini 4 Argon: Google's controlled-release model for long, complex knowledge and cyber tasks.
