AI in February 2026: agents arrive and the models leap

By Adam · Part of our monthly AI recap series, looking back at February 2026.

If January was about price, February was about capability. A wall of new frontier models landed, and the first genuinely useful agents arrived with them.

The models jumped. Anthropic shipped Claude Opus 4.6 (with a million-token context beta and agent teams) on 5 February, then Sonnet 4.6 — the new free-tier default — on the 17th. Google released Gemini 3.1 Pro, roughly doubling its predecessor's reasoning score.

Agents got real. Perplexity launched Computer, an agent that runs workflows for hours, and a Model Council feature that runs one query through three frontier models at once. Notion shipped Custom Agents in version 3.3 — multi-step autonomous workflows, free to try at first.

Also worth knowing: Google's Nano Banana 2 brought fast, high-quality image generation to the Gemini app and Lyria 3 added custom music tracks.

What it meant for you: the gap between "AI that answers" and "AI that acts" started closing. The tools you already paid for quietly got more capable — worth re-checking what your existing subscription could now do.

Full month-by-month detail lives in our AI Updates log. — Adam