Video Summary - Insane AI News Week: GPT-6 Astra, Fable 5.1, Gemini 3.8, NotebookLM Usage Limits....

Overview

This week’s AI news recap covers major model releases from multiple labs, along with notable product and usage changes.

GPT-6 Astra was announced as highly intelligent and aligned, with emphasis on:
- Computer use
- Professional knowledge work tasks
OpenAI highlighted a demo where users talk to ChatGPT while the system controls a computer in real time, appearing fast and accurate compared with earlier, slower workflows.
Benchmarks are described as extremely strong (“crushing” on paper), though the creator stresses they care most about real task performance.
Expectations: Astra may be able to agentically handle more work than prior models—especially for knowledge workers (e.g., research, documents, presentations, and potentially coding support).
Constraints / availability:
- At the time of recording, Astra is not for personal accounts.
- It’s rolling out first to certain businesses, with personal access expected shortly after.
- Pricing is expected to be as expensive as Anthropic’s Fable 5.1, making it less ideal for everyday use.
The OpenAI president is quoted suggesting Astra could be a “generational leap” toward AGI, framed as speculation until real-world validation.

Gemini 3.8 Flash was released with frequent updates to Flash models (3.7 Flash reportedly released ~3 weeks earlier).
Benchmark takeaways:
- In a general knowledge-work benchmark, 3.8 Flash scores poorly versus competitors.
- In narrower domains, it performs strongly:
  - Top for financial analyst tasks
  - Top for legal work
Value proposition: described as the cheapest Gemini option, meaning users may get more output for the same price—even if general knowledge-work quality isn’t best.
Availability:
- 3.8 Flash is already accessible inside the Gemini app and is suggested as the best Gemini choice right now.
- The creator criticizes Google for the ongoing lack of a new Gemini Pro model (Pro still at 3.1 Pro, despite earlier promises for updates).
Reasoning offered: many users don’t directly pick models; Google’s product surfaces (Search AI mode, Docs, Gmail features) often default to Flash because it’s fast and cheap, so Flash gets priority.

Fable 5.1 is positioned as an update that’s “better to work with” than the original Fable 5.
Claimed benchmark strength: described as best for knowledge work compared against other referenced models (including Claude Opus 5 and GPT-5.x).
The creator emphasizes quality-of-life improvements based on earlier complaints:
- Fewer false positives / guardrail friction
- More natural, human-friendly communication
  - less confusing verbosity
  - less need to ask it to “dumb it down”
- Lower price claim—but in the creator’s early experience it still seems roughly as expensive as before, or not noticeably cheaper.
Why excitement seems lower:
- Fable 5 hype declined after problematic availability and subscription/gating changes (including government-related access disruption and repeatedly shifting deadlines).
- Meanwhile, GPT-5.6 is portrayed as easier and more reliable, causing users to stick with it.
Main conclusion: the model you use isn’t always the “most intelligent,” but the one that’s easiest and least frustrating day-to-day.
The creator plans to test Astra vs. Fable 5.1 vs. GPT-5.6 soon, focusing on both intelligence and usability.

A disappointing update: Gemini Notebook now has usage limits based on compute consumption.
Previously, limits were more feature-specific (e.g., caps on generating certain media). Now there’s a broader compute-based cap across notebook outputs.
In settings, users can view:
- Current session limit
- Weekly limit
- Reaching 100% blocks further notebook use until reset
Reset cadence:
- Limits reset every 5 hours (previously daily resets).
- A weekly cap also applies.
Observed behavior / workaround:
- Within asset generators, there’s a meter showing estimated compute cost per output.
- Near the limit, there’s a “generate later” option that defers work until after the window—so it doesn’t consume the active 5-hour period.
Paid tier notes (as described):
- Pro ~4x free usage
- Ultra ~5x free usage
- The creator questions the pricing logic.
Other Notebook updates mentioned:
- Paid users can remove watermarks for generated images/videos.
- Rumor: possible native Chrome extension support, inferred from UI elements in an announcement video.

Quick platform updates include:

New plugin “Link” lets bots complete purchases using one-time credit cards (with user approval still required).
Bots can now be shared as templates, allowing others to create copies in their own accounts.
Users can select a voice per bot:
- The creator interprets this as a step toward real voice calling
- Calling isn’t available yet—only dictation.

The creator mentions rollout of Google Pix, previewed at Google I/O, with strong implications for image generation workflows.
A dedicated follow-up is planned once access is obtained.

A real-world incident involving three hikers in California:

AI can assist, but it can still be wrong. For potentially dangerous activities like hiking, rely on human judgment and experts.

This is framed as the “biggest AI week in a long time,” driven largely by model competition—especially among:

It also highlights friction and safety/accuracy concerns, including usage limits in Gemini Notebook and the hiking incident.

Presenter: The unnamed YouTube narrator (host of the weekly AI news recap)
Companies / cited contributors:
- OpenAI (including its president)
- Anthropic
- Google (Gemini Notebook team implied)
- Google Gemini (referenced as a system)
Sponsor mentioned: Explain by Scribba (Scribba / Explain)