Hey there, 

We’ve had some major AI news this week. Google released their new Gemini 3.8 Flash models while Anthropic released Fable 5.1. Perhaps the biggest news was OpenAI releasing their new GPT-6 Astra model which outperforms both Google and Anthropic. Here’s everything you need to know.  

Google shipped Gemini 3.8 Flash

On September 2nd, Google released their new Gemini 3.8 Flash model. The interesting thing about this inexpensive and powerful model is the pacing. It's only been three weeks since we got 3.7 Flash and we’re already getting another release. This marks the third Flash model we’ve gotten in six weeks which seems to indicate that Google is focusing their efforts on making fast, affordable models instead of Mythos-class competitors. Google initially announced that Gemini 3.5 Pro would be “coming soon” back at Google I/O but we still haven’t seen anything yet. 3.8 Flash is available for paid users across both the API and Gemini app. 

Gemini 3.8 Flash Cyber was also released on Wednesday. This model is only available for select cybersecurity teams and focuses on identifying vulnerabilities. 

Anthropic released Fable 5.1 and Mythos 5.1

Anthropic released their “most advanced model for coding and knowledge work” this week with Fable 5.1 and Mythos 5.1. These models perform better on both agentic and coding tasks which means Claude Code users will see a noticeable improvement in quality across reasoning levels. Perhaps more importantly, users will see a 25% drop in cost to complete the same task on Fable 5.1 as opposed to Fable 5 due to the improved reasoning. This push directly addresses one of the main concerns with Fable 5 and helps keep Anthropic near the top this week.  

Like previous versions, Mythos 5.1 will only be available to select partners while Fable 5.1 will be widely available. Paid users can already start working with Fable 5.1 on all platforms. 

OpenAI launched GPT-6 Astra and called it AGI

Early on Thursday, OpenAI dropped their highly anticipated new model GPT-6 Astra. GPT-6 comes with some significant performance improvements, mostly around agentic work, coding, mathematics, and cybersecurity. On one notably difficult benchmark, the model was able to get the first ever perfect score and even outperform Anthropic’s Fable 5 model. This new flagship model is currently available to a select group but will be rolling out to paid users in the next few days. The framing around GPT-6 has been dominated by the comments made by OpenAI’s president, Greg Brockman, who said “welcome to the AGI era”. While GPT-6 Astra appears to be one of, if not the best, models in the world, I wouldn’t consider it actual AGI. Nevertheless it seems like these comments have sparked considerable excitement about OpenAI’s future. 

My Thoughts

This was one of the most exciting weeks of 2026 for the AI community. We got new models from three of the biggest frontier labs, each pushing the boundary further. Google gave us better performance on their inexpensive Flash models, Anthropic gave us a more affordable upgraded Fable 5.1, and OpenAI gave us their new flagship model. The claims of an AGI era seem a bit overhyped but definitely illustrate just how far these models have come in the past few years.

Google’s lack of a Gemini Pro release reinforces the idea that they’ve shifted their focus to affordability and speed for now and will give us a new Pro model when it can outperform Fable and Astra. I’m excited to use GPT-6 within tools like ChatGPT Work. This was already a powerful feature last week but now I have a feeling it's going to be one of the best AI tools that are publicly available.