Gemini 3.1 Pro, 16k tokens/second, & AI for the Dead

TLDR;

Welcome back! This week, we take a look at new models and product releases, especially Gemini 3.1 Pro; new billion-dollar investment rounds; new hardware that spits tokens at 16,000 per second; the curious case of Korean stocks, crucial to AI but “ignored” by the rest of the world; and the latest example of a company that might try to use AI to bring people from the dead.

Enjoy!

THEWHITEBOX

Google Drops Gemini 3.1 Pro

Google has released Gemini 3.1 Pro as an upgraded “core intelligence” model. It said rollout starts immediately across multiple surfaces:

for developers in preview via the Gemini API in Google AI Studio (and also via Gemini CLI, Antigravity, and Android Studio),

for enterprises via Vertex AI and Gemini Enterprise,

and for consumers via the Gemini app and NotebookLM.

On performance, Google highlighted improved reasoning and reported a verified 77.1% score on ARC-AGI-2, describing that as more than double the reasoning performance of “3 Pro,” and superior to all other models available today.

Here’s an example of a very impressive use case.

It frames the model as “intelligence applied,” pointing to examples such as generating code-based animated SVGs from text prompts, synthesizing complex systems into dashboards (including an ISS-orbit visualization from public telemetry), building interactive 3D experiences, and translating literary themes into functional website code.

TheWhiteBox’s takeaway:

Overall, the reception seems to be “good” but not “great”, with some experts calling it a “previous generation model” compared to others like GPT-5.3 Codex or Claude Opus 4.6.

However, it seems to have improved in meaningful ways. For instance, the model does seem to have improved hallucinations tremendously, which were a very relevant issue with past models, at least according to Artificial Analysis, dropping from 88% to 50%:

It also takes the top spot in the overall intelligence index:

Naturally, as I always say, benchmarks only take you so far, and real usage will determine whether this model is worthy of the position benchmarks put it in.

Specifically, the key is whether this model can make Google’s coding products become as useful and used as Claude Code or Codex from Anthropic and OpenAI, respectively, which seems to be the killer use case, at least in terms of growing revenues.

THEWHITEBOX

Rork Launches Rork Max

The start-up Rork has introduced Rork Max, a platform focused on creating Apple apps that look absolutely incredible.

It’s basically a vibe-coding app, similar to Claude Code or Codex, but focused exclusively on mobile applications, allowing you to create games and other apps not only for iPhone but also for the Apple Watch, iPad, and even Apple TV.

The video above is not just a teaser but an actual walkthrough (more or less) of how one of the team members builds a copy of Pokémon Go. Quite special.

TheWhiteBox’s takeaway:

Remember when I was saying just two weeks ago that the acceleration was coming? Well, here’s another example for you.

Again, this doesn’t promise full-stack apps in one shot, but always think not of…

Enjoyed the introduction? Subscribe or sign in to continue with this edition.