Weekly AI Recap — Gemma 4 12B, MiniMax M3, The AI IPO Wave | June 2026

Weekly AI Recap — Gemma 4 12B, MiniMax M3, The AI IPO Wave | June 2026

More

Summary

Fahd Mirza’s weekly AI recap for the first week of June 2026 covers a dense stretch of model releases, tool launches, and an IPO wave that could mark a turning point for the AI industry. On the model side, highlights include Google DeepMind’s Gemma 4 12B unified architecture (256K context, Apache 2 license, no separate encoders for image and audio), Ideogram’s first open-weight 9.3B text-to-image diffusion transformer, Nvidia’s Locate Anything visual grounding model with parallel bounding-box decoding, Cosmos 3 omnimodal world model, JetBrains’ Mel 2 mixture-of-experts coding model (12B total parameters, 2.5B active, 131K context), and MiniMax M1 — described as the first open video model combining frontier-level coding, a 1M token context window, and native multimodal understanding.

On the tools side, Siri Search shipped a native desktop app for the Hermes agent across Mac, Windows, and Linux, and OpenAI quietly launched instant live-hosted web apps directly from Codex natural language prompts with no deployment configuration required.

The episode’s headline story is a simultaneous three-way IPO push: Anthropic confidentially filed its S-1 with the SEC, reporting a $47 billion monthly revenue run rate; OpenAI filed its own confidential S-1 targeting a September Nasdaq listing at a valuation between $852 billion and $1 trillion, with Goldman Sachs, Morgan Stanley, and JP Morgan leading; and SpaceX targeted a $1.77 trillion valuation with first trading under ticker SPCX aimed at June 12th.


📺 Source: Fahd Mirza · Published June 06, 2026
🏷️ Format: Roundup

1 Item

Channels

5 Items

Companies