What were the biggest AI announcements from 28 September to 1 October 2026?

Google released Gemini 4 Argon to trusted cyber defenders on 30 September, OpenAI launched always-on Dots agents at DevDay on 29 September, Anthropic released Claude Sonnet 5.5 on 28 September, and NVIDIA announced its Open Agent Safety Platform with OpenShell and Sentry on 28 September.

Is Gemini 4 Argon available to everyone?

Not yet. Google is rolling Gemini 4 Argon out first to trusted cyber defenders through its Fairwind Program, with wider availability to paid API customers and Google AI Ultra subscribers described as coming soon. Launch pricing is $2 per million input tokens and $10 per million output tokens.

How much does Claude Sonnet 5.5 cost?

Token prices are unchanged from Sonnet 5 at $2 per million input tokens and $10 per million output tokens. Anthropic says Sonnet 5.5 runs more than 30% faster and costs up to 30% less for most work because it needs fewer tokens for the same task.

What are NVIDIA OpenShell and Sentry?

OpenShell is NVIDIA's open-source secure runtime that limits what files, system calls and tools an AI agent can use. Sentry is a watchdog running on NVIDIA BlueField-4 DPUs that monitors agent behaviour and enforces policy in hardware. Together they form the Open Agent Safety Platform announced on 28 September 2026.

What's New in AI: Gemini 4 Argon, OpenAI Dots, Claude Sonnet 5.5 and NVIDIA's Agent Safety Platform

Week of Sept 28 – Oct 1, 2026 · Published October 1, 2026 · Compiled by SCR

Four launches defined the last week of September 2026. Google released Gemini 4 Argon to trusted cyber defenders first, OpenAI introduced always-on Dots agents at DevDay, Anthropic shipped Claude Sonnet 5.5 at unchanged prices, and NVIDIA released OpenShell and Sentry to keep AI agents inside their limits.

Model

Google's Gemini 4 Argon goes to cyber defenders before anyone else

Google announced Gemini 4 Argon on 30 September 2026 as its new frontier model for real-world software engineering, enterprise knowledge work such as legal and finance, and cyber defence. It is rolling out first to trusted cyber defenders through Google's Fairwind Program, with a wider release to paid API customers and Google AI Ultra subscribers described only as coming soon.

Launch pricing is $2 per million input tokens and $10 per million output tokens, rising to $4 and $20 after the introductory period. Google says the output limit grows to 1 million tokens, up from 64,000, and reports 77.9% on the DeepSWE v1.1 software engineering benchmark.

TakeHanding the unguarded version to a vetted list of defenders first is becoming the release pattern for frontier models. The defenders get a head start; everyone else gets the version with the guardrails on.

Sept 30, 2026 Source: Google

Agents

OpenAI launches Dots, always-on agents inside ChatGPT

OpenAI's DevDay headline on 29 September 2026 was Dots: always-on agents that pursue goals you set with little supervision. Each Dot runs on the GPT-6 Astra model and is available in ChatGPT and Codex for Pro and Business Premium subscribers in eligible markets.

Dots work through Slack and Microsoft Teams, with text messaging to follow, and Microsoft's Agent 365 security controls are integrated. OpenAI says that over time it envisions teams of Dots working together on a user's behalf.

TakeAn agent that works for you around the clock also holds your logins around the clock. The security controls are the part of this launch worth reading first.

Sept 29, 2026 Source: TechCrunch

Model

Anthropic ships Claude Sonnet 5.5: faster, and cheaper per task at the same price

Anthropic released Claude Sonnet 5.5 on 28 September 2026, the second model in its Claude 5.5 family. It runs more than 30% faster than Sonnet 5 and, because it typically needs fewer tokens for the same job, costs up to 30% less for most work, while token prices stay at $2 per million input and $10 per million output.

Anthropic reports 70.6% on Terminal-Bench 4.0 for agentic coding and says knowledge-work results come close to Opus 5.5. It is available on Claude.ai, the Claude Platform, Amazon Web Services, Google Cloud and Microsoft Azure, with cyber safeguards comparable to Opus 5.5.

TakeSame price per token, fewer tokens per job. For anyone budgeting AI workloads, cost per finished task is now the number to track, not the rate card.

Sept 28, 2026 Source: Anthropic

Safety

NVIDIA's OpenShell and Sentry put AI agents inside a cage

NVIDIA announced its Open Agent Safety Platform on 28 September 2026, an open reference design for securing AI agents from testing to deployment. OpenShell, an open-source secure runtime, runs an agent without privileges, limits the files it can reach and filters its system calls in the kernel, routing requests through a single approved channel.

Sentry adds an out-of-band watchdog on NVIDIA BlueField-4 DPUs that monitors agent behaviour and enforces policy in silicon, invisible to the agent itself. In Vera Rubin POD systems, BlueField-4 sits on the node's only path to the model.

TakeAgent safety is moving from the model into the hardware. The same week OpenAI shipped agents that run around the clock, NVIDIA shipped a watchdog that sits where they can't reach it.

Sept 28, 2026 Source: NVIDIA

SCR Update

Our AI Build Configurator adds AMD Instinct and GB200 NVL72

Version 1.4.0 of the licensed AI Build Configurator, released on 1 October 2026, adds five AMD Instinct accelerators from MI300X to the Helios MI455X rack and GB200 NVL72 as its own option. It also reprices GB300 NVL72 by 26% and models Virginia's new $0.011/kWh data centre electricity tax. Full details are in the v1.4.0 release notes.

Oct 1, 2026 Release notes