Gemini 4 Argon launches at $2 per million input tokens, Replit moves its default to OpenAI, open-weight models take most gateway tokens, Norway fines Xandr, and
By Avik Ghosh, Managing Founder, illuminis
Five stories from the past week that matter to anyone buying, running or regulating AI, with our take on each.
On October 13, Google made Gemini 4 Argon generally available to developers and enterprises through the API in AI Studio and the Gemini platform. Google launched it at an introductory price of $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% off, and said it ranks first on the Vals Index, which measures performance in finance, coding, legal and tax work. Source.
Our take: That is the same list price OpenAI set for GPT-6 Sol in September, so frontier models are now competing head to head on price. The word "introductory" matters: a price that can change is a poor basis for a long-term commitment to one model. Pick the model that finishes each step at the lowest total cost, and be ready to switch when prices or results move. That is the case for CompletionPrism™.
On October 15, speaking at TechCrunch Disrupt, Replit CEO Amjad Masad said the company had made OpenAI its default model provider in place of Anthropic, citing cost, partnership and performance. He said Replit had until recently been one of Anthropic's top five customers. Claude models remain available to Replit customers who select them, and TechCrunch noted that most customers stick with the default. Source.
Our take: Replit could make this change because its product was built to work with more than one model. A software vendor wired to a single provider has no such choice when costs rise or a competitor pulls ahead. Being model agnostic is now part of the product, and part of the sales story. See the Embedded License.
Vercel data reported on October 17 shows open-weight models processed more than seven in ten tokens on its AI Gateway in September, and their share of spending rose from about one in twenty dollars in June to nearly one in four. The average price per token was about 47% lower than in December 2025, and almost one in three teams changed their most-used AI provider in September. Claude Opus 5.5 took the largest share of spending at 11.03%. Source.
Our take: Teams are already sending routine work to cheaper models and saving premium models for the steps that need them. The same report notes that China has proposed export controls on AI items, which adds supply risk to some of those cheaper models. Choosing the model for each step, with a fallback ready, is now a cost decision and a risk decision. See CompletionPrism™.
In a decision dated October 2 and published by the European Data Protection Board on October 16, Norway's data protection authority fined Xandr, now part of Microsoft, €8.6 million for processing data about Grindr users during real-time ad bidding without a lawful basis. Xandr relied on consent collected by Grindr, which the authority found invalid. The authority held that accountability cannot be outsourced and that each joint controller must check for itself that consent is valid and covers its own processing. Source.
Our take: Any marketing team that buys or enriches contact data on a partner's word is in the same position as Xandr. If you cannot show where a contact came from and why you may use it, someone else's consent will not protect you. That is why RelationLens shares a contact only when the employee approves it, and records that approval.
On October 14, the US Federal Maritime Commission proposed a fee of $200,000 per voyage on Spanish-flagged vessels and possible limits on imports of Spanish goods, including perfumery and cosmetics, oils and fats, ceramics, stone articles, and sugar and confectionery. The FMC said Spain had denied US-flagged vessels access to its ports. Public comment closes on December 14. Source.
Our take: Buyers of these categories have about eight weeks before a decision that could change their landed costs or cut off a source. Now is the time to map which suppliers and lanes are exposed and line up other sources, before the rule is final. That is the work SourceMind™ does.
In the News comes out every week. Subscribe from the Insights page to get it by email.