Blog / Topic
Language, image and multimodal models: what they can do and how they change.
Claude Fable 5.2 just took first place on a major AI leaderboard, at four times the runner-up's price. Why its lead is shakier than it looks, and the few people who should actually pay.
Read the storyFive AI companies raced new models out between 23 October and 9 November. One crossed the line first, one improved the most for less money, and one barely moved. Here's the results sheet.
Gemini 4 Carbon tops a public vote where people pick the answer they like, then slips to second on a stack of hard tests. Why the lists disagree, and why neither win is as solid as it looks.
A model called Krypton is trending on X as Google's next Gemini. Google hasn't announced it, and there are no scores or prices. What the rumour says, and what to do until it's real.
The name gpt-6.2-astra has turned up in people's Codex usage data, and a launch was tipped for 6 November. OpenAI still hasn't announced it. What's confirmed, what's rumour, and why the last Astra never shipped.
Elon Musk says xAI's next model, Grok 4.8, is enormous and built on new, faster training software. There's no price, no test results and no release date yet. What the claim means in plain English, and what to do until it's out.
A simple method for choosing an AI model for coding, writing or data analysis. Start cheap, test on your own work, and only pay more when the cheap one fails.
The strongest AI model you can download and keep is Kimi K3.1 Pro. It ranks #16, and renting it costs as much as a closed model that beats it. The real bargain sits six places lower.
