Blog

LFM3 is Liquid AI's new family of six open 2B-class models for text, vision, audio and decisions. What each one does, how fast it runs, and where it falls short.

Genie 4 is Google DeepMind's new world model with native audio and Gemini in the loop. What changed from Genie 3, who can use it, and what Google left out.

GPT-6.1 Terra launched on 26 October 2026 at $1/$5 per million tokens, half of Sol. It is built for screens, forms and writing. What changed and who it fits.

A Canva Craft review from an engineer who ships image-model features: how the plan-then-render design works, where it ranks, what it costs, and who should use it.

Canva Craft is Canva's first in-house image model. It plans the layout before rendering at 2K, so it handles posters and text well. Ranks, access and cost inside.

Seven Kimi K3.1 Pro alternatives, priced per 1,000 requests and per finished task: Claude Sonnet 5.5, Opus 5.5, Gemini 4 Argon, Gemini 3.8 Flash, DeepSeek and more.

Block Dessa is a full-duplex voice model trained only on recorded human speech. What Block has shown, what it hasn't, and whether it matters for support teams.

My Kimi K3.1 Pro review: a 3.5/5 model that thinks close to Claude Opus 5.5 at $2/$10 per million tokens, but runs slow, writes long, and has no weights yet.

Kimi K3.1 Pro pricing is $2/$10 per million tokens on the API, or $19 to $199 a month on Kimi plans. What a real task costs, and where the bill grows.