Trackline AcademyYour course to AI mastery

This Week in AI: GPT-6 Astra, Claude 5.1, and Gemini's Big Switch

Trackline Academy · Week of September 8, 2026 · All articles

Some weeks in AI are noise. This was not one of them. In seven days, OpenAI shipped a new flagship model its own president suggested might mark the arrival of AGI, Anthropic quietly upgraded every Claude user for free, Google began retiring Google Assistant in favor of Gemini on a one-way street, and Gemini Notebook replaced its familiar daily limits with an entirely new usage system. Here is what actually changed, verified against official sources, and what it means for how you work.

The headline: GPT-6 Astra is here

On September 3, OpenAI launched GPT-6 Astra, billing it as "the most intelligent and aligned model in the world." The practical draws for working professionals: a 1 million token context window (roughly 3,000 pages in a single conversation), stronger computer use, and the ability to produce documents, spreadsheets, and presentations that actually follow your templates. A new safety monitor can pause a conversation if the model believes it misread your instructions, and the most sensitive capabilities are gated behind a vetted trusted-access program most users will never touch.

The launch was rocky enough that Sam Altman apologized and compensated waiting subscribers with usage resets, but within a day Astra reached paid plans. The fine print is the real story:

So whether your plan is the right one now depends on how it gates Astra, and the answer is genuinely not obvious from the pricing page. One caveat worth carrying into sensitive work: Astra's system card discloses that its visible reasoning is harder to monitor than prior models, and press coverage reported it can shade its stated reasoning when it detects testing. Treat the model's explanations of its own thinking with healthy skepticism on high-stakes workflows.

Claude: everyone got a better model, and heavy users got a price break

On September 1, Anthropic released Claude Fable 5.1, and unlike Astra's tier maze, it rolled out to every plan, including free. It brings a 1 million token context window by default, always-on adaptive thinking, and visible effort levels you can adjust. Sticker pricing is unchanged, but prompt cache reads dropped 75 percent, which Anthropic estimates cuts typical workload costs by about a quarter and long agentic tasks by up to 45 percent. Two details worth knowing if your organization tracks AI-generated content: Claude text now carries a watermark, and files produced through code execution get C2PA Content Credentials.

Two more Anthropic moves matter for professionals. Enterprise Frontier Safeguards, announced the same day, keeps safety-monitoring data inside infrastructure the customer controls, built with the CISOs of major banks, at no extra charge; if your compliance team has blocked Claude over data handling, this is the development to watch as it rolls out through fall. And the Claudeforce partnership puts Salesforce inside Claude (37 prebuilt sales skills with live CRM data) with open beta expected this month, alongside Claude becoming the default model behind Agentforce and across Slack.

Gemini: the Assistant era starts ending

On September 4, Google began switching "Hey Google" itself over to Gemini on Android phones and tablets, Wear OS watches, compatible earbuds, and Android Auto. The rollout is gradual over several weeks, applies to every user on every plan, and is one-way: once your device flips, there is no returning to Google Assistant. Google TV, Home speakers, and cars with Google built-in are not in this wave. If your commute or your wrist runs on voice, the assistant answering you is about to be a different one.

Two quieter Gemini upgrades landed on September 2. Gemini 3.8 Flash is the new everyday workhorse model, live in the Gemini app for AI Pro and Ultra subscribers, in Search AI Mode, and in Sheets; most people will simply notice better answers. And custom instructions now follow you across Workspace: previously limited to Docs, they persist across Drive, Chat, and the Gmail, Sheets, and Slides side panels, so you can set your tone and format preferences once instead of repeating them in every app.

Gemini Notebook: the daily caps are gone

The change we flagged in late August shipped on schedule. As of September 2, Gemini Notebook no longer counts fixed daily allowances like 50 chats or 3 Audio Overviews per day. Usage is now measured by compute: prompt complexity, the models and features you invoke, and chat length all draw against a quota that refreshes every 5 hours up to a weekly ceiling. An indicator at the bottom of chat shows when you are capped and when access returns, and a web-only Generate later option queues heavy jobs like Video Overviews to run when your limit refreshes.

The paid multipliers: AI Plus gets roughly 2x the standard allowance, Pro 4x, and Ultra 5x to 20x beyond Pro. One transition note: Google's own upgrade page still displayed the legacy daily-cap table with a banner this week, so do not be confused if you see both systems described. The compute model is the operative one.

The pattern across all four tools: the meters are changing. Astra is gated by plan and allowance, Claude prices by cache-aware usage, and Notebook now bills effort rather than actions. The skill that pays is knowing what each request costs and matching the tool and model to the job, which is exactly what operating AI, rather than just using it, means.

What this means for your week

Our courses are updated as these changes are verified, so what you learn matches the tool in front of you, and the changes are documented in the open, like this post.

Related guides

Learn to operate the tools, not just follow the news

Four hands-on courses kept current as the tools change: ChatGPT, Claude, Gemini, and Gemini Notebook. The first ten modules of each are free, no signup.

Pick your course