This observation covers changes formally announced or reliably confirmed between 08:36 on 30 September and 08:36 on 1 October 2026 in Australia/Sydney. Two items are included.
Gemini 4 Argon enters limited access for trusted cyber defenders
Google announced Gemini 4 Argon on 30 September and began providing access to a group of trusted cyber defenders through the Fairwind Program. It is not currently a public API for developers, enterprises or consumers, and Google has not provided a date for broad availability. Google says Argon is designed for long-running workflows in software engineering, finance and legal knowledge work, and cyber defence, with its output limit increased from 64,000 tokens to one million tokens. Results published for DeepSWE v1.1, AutomationBench, the Vals Index and cybersecurity evaluations are developer-reported. Reuters independently confirmed the model name, limited-access status and absence of a public-release timetable, but did not independently reproduce the benchmark results.
Google also says Argon is participating in the US government's voluntary pre-release access process. Wider access for paid API customers and Google AI Ultra subscribers remains a later step. The announced introductory prices of US$2 per million input tokens and US$10 per million output tokens, together with the cached-input discount, describe the intended public launch rather than a service that the general public can currently purchase. Google's safety description covers harmful-request refusals, indirect prompt-injection defences, chain-of-thought and action monitoring, and sandbox isolation. The public material does not provide a complete system card, test samples or external production failure rates.
Direct sources: Google, Gemini 4 Argon: our next era of frontier intelligence, 30 September 2026, https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/; Google DeepMind, Gemini 4 Argon model page, 30 September 2026, https://deepmind.google/models/gemini/; Reuters, Google announces Gemini 4 flagship AI model after months of delays, 30 September 2026, https://www.reuters.com/legal/litigation/google-announces-gemini-4-flagship-ai-model-after-months-delays-2026-09-30/
CoreWeave opens limited NVIDIA Vera Rubin NVL72 production capacity
CoreWeave announced on 30 September that NVIDIA Vera Rubin NVL72 was in limited availability on its cloud platform, with hundreds of Rubin GPUs deployed across multiple regions. Cognition became the first customer to run production agent workloads on the system, using it for Devin software-engineering inference, training and reinforcement learning. This series previously recorded scaled Vera CPU shipments and the first server delivery to AWS. The new fact is that the NVL72 rack-scale system, combining 72 Rubin GPUs, 36 Vera CPUs and NVLink 6, has reached a cloud production customer rather than remaining on a roadmap or in engineering testing.
Cognition reports that its SWE-2 inference workloads achieved up to 4.8 times the total token throughput per GPU of GB200 NVL72 at matched interactivity, while reinforcement-learning output throughput improved by 3.8 times. Cognition engineers ran the tests in CoreWeave's environment. Barron's independently confirmed platform availability and Cognition's production use, but the performance figures still come from the supplier and its first customer, with no reproduction by a disinterested third party. CoreWeave explicitly describes the current status as limited availability, so it is not yet broadly available or deployed at general scale.
Direct sources: CoreWeave, Cognition Becomes First Customer for NVIDIA Vera Rubin NVL72 on CoreWeave Cloud, 30 September 2026, https://coreweave.com/blog/cognition-becomes-first-customer-for-nvidia-vera-rubin-nvl72-on-coreweave-cloud; Barron's, CoreWeave Stock Rises After Neocloud Goes Live With New Nvidia Hardware, 30 September 2026, https://www.barrons.com/articles/coreweave-stock-nvidia-neocloud-ai-f651e463
Sources
https://blog.google/innovation-and-ai/models-and-research/gemini-models/gemini-4-argon/
https://deepmind.google/models/gemini/
https://www.reuters.com/legal/litigation/google-announces-gemini-4-flagship-ai-model-after-months-delays-2026-09-30/
https://coreweave.com/blog/cognition-becomes-first-customer-for-nvidia-vera-rubin-nvl72-on-coreweave-cloud
https://www.barrons.com/articles/coreweave-stock-nvidia-neocloud-ai-f651e463
Discover more from Geoffrey Chen
Subscribe to get the latest posts sent to your email.