Gemini 4 Argon: announced, but not broadly available
Google announced Gemini 4 Argon on September 30, 2026, with initial access through its Fairwind program for trusted cybersecurity defenders. Google also says it is participating in the U.S. government’s voluntary pre-release access process. Wider distribution is intended to start with paid API customers and Google AI Ultra subscribers, but the announcement provides no public release date. An announced model is not yet an available development dependency.
Google lists introductory pricing of $2 per million input tokens and $10 per million output tokens. Its footnote establishes an important budgeting detail: those rates will become $4 and $20 respectively after the introductory period. The company also specifies a one-million-token output limit—not merely a statement about input capacity.
Google reports 77.9% on the DeepSWE v1.1 software-engineering benchmark and describes AI-assisted C/C++-to-Rust migrations. Crucially, it says critical rewrites undergo automated and manual auditing, emulation testing, and review before production deployment. These are company-reported results, not independent confirmation of performance across customer workloads.
The practical interpretation is straightforward: evaluate the model when access arrives, rather than treating a benchmark position as a procurement verdict. Google’s own account also argues against the tempting shortcut of letting impressive generated code bypass engineering review.
Xbox’s reset includes a boundary around AI
Michael Parekh’s AI: Reset to Zero newsletter frames Xbox’s transition around a New York Times profile of Asha Sharma. Parekh’s account describes restructuring alongside ambitions for cloud expansion, Minecraft investment, and another console generation. That is the newsletter’s attributed account of the profile; it should not be mistaken for a complete hardware roadmap.
Microsoft’s primary announcement supplies firmer context. Sharma became EVP and CEO of Microsoft Gaming on February 20, 2026. An April 23 update says the organization is now called XBOX and her title is CEO XBOX. Microsoft also promoted Matt Booty to chief content officer, while Phil Spencer planned an advisory handoff through the summer.
Sharma’s opening commitments explicitly paired renewed console investment with expansion across PC, mobile, and cloud. Her message also rejected pursuing short-term efficiency or filling the ecosystem with “soulless AI slop,” emphasizing human-created games.
That last point complicates any simple “AI executive takes over, therefore AI replaces game development” narrative. Microsoft’s stated direction supports technological innovation, but puts creative quality ahead of indiscriminate automation. For Windows players, the useful test will be whether cross-device ambitions produce a better PC experience—not simply more platforms on a presentation slide.
AI safety: agreement does not erase disagreement
The public development is a voluntary accord. AP reports that President Donald Trump and technology leaders signed it on September 29, 2026. Its measures include internal controls, an independent external auditor, and a board committee to review auditor reports. AP also notes that some measures resemble existing company practices or earlier commitments.
There is evidence of continuing tension beneath that agreement. In Wall Street Journal reporting distributed by Dow Jones, executives including Nvidia’s Jensen Huang privately questioned Anthropic’s Dario Amodei about his public risk warnings. According to people familiar with the discussions, Amodei defended being candid about capabilities and dangers.
The same report says a FINRA-style self-regulatory organization was abandoned after Huang, Mark Zuckerberg, and Elon Musk raised concerns about concentrating power in leading AI companies. That supports a narrower conclusion than universal movement toward a political “middle”: the companies negotiated shared principles while retaining disagreements over warnings and oversight.
Separately, Dow Jones reports that Greg Brockman told OpenAI employees he and his wife would not fulfill a second $25 million pledge to Leading the Future. The reported employee message was sent in June; it was not necessarily a decision made during this week’s negotiations. A personal political contribution decision does not, by itself, establish a change in OpenAI’s model-safety policy.
Kindle turns the page more literally
Amazon’s refresh covers Kindle, Paperwhite, and Colorsoft, with thinner designs and new accessories. TechCrunch reports that the basic Kindle was available in the United States on October 1, while Paperwhite and Colorsoft were available for preorder.
Amazon’s Australian announcement confirms a useful compatibility distinction: Kindle Click, its Bluetooth page-turning remote, supports new devices and Kindle models released in 2024 or later. The physical-button Page-Turn Cover is for the new Paperwhite Signature Edition and Colorsoft Signature Edition—not every reader in the family. Those compatibility details should not be confused with Australian prices or regional shipping schedules.
The broader takeaway is less theatrical than “everything has changed,” but more useful: distinguish limited access from availability, strategic promises from shipped experiences, voluntary oversight from resolved safety disputes, and accessory announcements from compatibility. Turning a page is easy. Knowing what is printed on the next one takes more work.
Update: Gemini 4 Argon reportedly opens beyond restricted access (October 16, 2026)
PYMNTS reports that Google opened Gemini 4 Argon to developers, businesses, and paying consumers on October 13, moving beyond the Fairwind-only access described earlier. If confirmed in Google’s own availability documentation, that would make the model an option for real-world evaluation rather than a restricted preview.
The report also says the introductory $2-per-million-input-token and $10-per-million-output-token rates will double after November 30. It attributes to Google a first-place result on the Vals Index for finance, coding, legal, and tax work, and says Wiz has used the model to identify and patch security flaws affecting sensitive records.
References
- Meta Raises Its AI Game as Google Opens Gemini 4 Argon - PYMNTS.com PYMNTS.com · 2026-10-15T23:03:41+00:00
- ‘Turning the Page’ in AI. Google, Microsoft & AI Safety. ARD #175 - AI: Reset to Zero AI: Reset to Zero · 2026-10-01T20:03:44+00:00
- The new Kindle ditches the raised bezel in a push toward a smaller, lighter e-reader | TechCrunch techcrunch.com