Are you tired too? Woke up in the morning — OpenAI rolled out an agent. Went for coffee — Anthropic released a new model. Sat down to work — Google dumped a ton of announcements from I/O 2026.

Well, let's at least talk about the most important things from the last I/O.

1️⃣ Gemini 3.5 Flash: Cheaper, faster, meaner
Everyone already understands that running heavy LLMs (Pro/Ultra level or GPT-5) for every little thing is madness and budget burning. Google is betting on fast models and dropped Gemini 3.5 Flash.
Its essence is an ultra-fast and cheap grinder for long-horizon tasks. They claim 76.2% on Terminal-Bench 2.1 (outperforming competitors in code writing). The future is no longer about one smart prompt, but about a crowd of cheap sub-agents that solve a decomposed task in parallel and check each other.

2️⃣ Gemini for Science — the end of manual brute force for DS
A specialized toolset for researchers and Data Science. They've integrated with 30+ databases (AlphaFold, UniProt), but the main thing here is the AlphaEvolve project. This is an agentic engine that simultaneously generates and scores thousands of code variations to test hypotheses. Perhaps computational experiments or basic EDA will soon be entirely handed over to a script for a few cents.

3️⃣ Gemini Omni — video as native output
Native multimodality has reached video. The Gemini Omni model consumes text, code, audio, and images, and spits out video, understanding the physics of the process. You can edit video simply with text prompts in chat mode, maintaining character consistency.

4️⃣ Paywall for the IT elite
The warming up of the elites continues. Google rolled out the AI Ultra plan for $100 per month. Specifically for devs: priority access to Antigravity, 5x request limits, and 20 TB of cloud storage for your datasets. 100 bucks a month for an IDE with a neural network attached.

We keep earning for endless subscriptions to keep earning 🫠