This website uses cookies
Read our Privacy policy and Terms of use for more information.
DeepSeek prices cache-hit tokens at $0.006 per million, OpenAI stops selling $200 Pro seats with no reopen date, and a community benchmark finds the same model costs 17x more per correct answer depending on what you wrap it in.
Google's Gemini intro rate expires December 31 and doubles January 1, GPT-6 Astra lands at 2.5 times Sol's input price, and a 380-category study found that 59.8% of the citations behind Perplexity's software recommendations point at domains ranked worse than #100,000.
OpenAI burns $65 of compute on a $20 subscription while an AI takedown agent keeps an app off Google Play for 46 days, and Gemini flags an authentic product as counterfeit
Nvidia moved Claude Opus 5 from 30% to 100% on the same benchmark without changing the model. Linear's data on 199,000 users shows agent teams tripled their output and saved nobody any time. The lever moved to the layer around the model, and almost nobody has that layer on the P&L.
Gemini's app passed 1 billion monthly users, with small businesses generating 150 million images a day on it. Days later Google made the visible watermark optional. And research confirmed the newest reasoning models hallucinate more than the ones they replaced.
One Rippling engineer was burning $50,000 a month on tokens. Shopify's AI-driven orders tripled. A German tax firm counted 40,000 hours of recovered capacity. The measuring has started, and the numbers cut in both directions.
OpenAI cut GPT-5.6 Luna by 80% and DeepSeek V4 Flash landed at $0.14 per million input tokens. In the same week, researchers reproduced a self-spreading Copilot worm that silently halves the numbers in your financial documents.
Anthropic held Opus 5 at $5 per million and doubled its benchmark score. Then it named 49,000 accounts it says strip-mined Claude to build the cheap alternatives. Here's what that pairing tells you about the models on your invoice.
Anthropic reversed its own Fable 5 pricing plan under competitive pressure. NYC just told landlords to disclose AI-altered photos. Here's what backing down signals about where AI actually stands.