DeepSeek raises prices while US labs cut + IBM picks OpenAI

DeepSeek's own pricing page shows new peak and off-peak rates taking effect on August 16: peak output tokens for its V4 Pro model climb from 87 cents per million to three dollars and ninety-six cents, and some cached-input rates rise roughly 1,100 percent. OpenAI and Anthropic are moving the other way, cutting GPT-5.6 Luna by as much as 80 percent and scrapping a planned September increase for Sonnet 5. TechRepublic, citing SiliconData figures reported in the Financial Times, says DoorDash, Airbnb and Coinbase now run Chinese models in production.

IBM said on August 13 that GPT-5.6, Codex and ChatGPT Work will be embedded in IBM Consulting Advantage, the platform its consultants deliver client work through. The company is standing up a dedicated OpenAI practice with thousands of certified consultants and engineers, starting with financial services, government, telecommunications and retail. IBM also joins OpenAI's Daybreak Cyber Partner Program, pairing the models with its Autonomous Security service. No financial terms were disclosed.

A new Media Watermark toggle in settings lets users switch off the visible mark Google stamps on output from Gemini, the Nano Banana, Omni and Lyria models, and the Flow video editor, with Search support to follow. The invisible SynthID watermark and C2PA provenance metadata stay on either way, and Google open-sourced a library called Credentio so developers can validate C2PA credentials locally. The change arrives two weeks after the EU AI Act transparency duties and California's disclosure deadline became enforceable.

In Elliott versus New York Bariatric Group, Judge Walter M. Spader Junior found that the self-represented plaintiff had buried instructions in tiny white text telling any artificial intelligence system reviewing the document to rule in his favor. The tell was odd stretches of white space in the filings. Spader revoked the plaintiff's electronic filing privileges, requiring paper submissions in person, and wrote that he could find no Connecticut or other United States decision squarely addressing the conduct.

Qwen3.8-27B landed on Hugging Face with open weights, no gating, and native image and video input. Its model card reports 61.7 on SWE-bench Pro and 84.3 on the OSWorld computer-use benchmark, with a native context window of 262,144 tokens that extends to one million. The license matters as much as the scores: the flagship Qwen3.8-Max shipped days earlier under custom terms carrying a $50 million revenue trigger, while this one is unrestricted Apache 2.0.

HEIR, short for homomorphic encryption intermediate representation, converts a pretrained model so that it computes on encrypted inputs, meaning the server holding the model never sees the underlying data. Google published the compiler on GitHub alongside four working demonstrations built with outside partners: a recommendation model, credit-card fraud detection, network intrusion detection and hotword detection, each timed on a single processor thread. Homomorphic encryption has been sound in theory and too slow in practice for years, and latency is the half this targets.

The two companies expanded an existing partnership on August 14, building out from the commercial service they run in Zagreb with local operator Verne to four more European cities that neither has named. Pony.ai supplies the Level 4 autonomous driving stack, Uber supplies booking and payments, and local partners run the fleets day to day. A Middle East deployment is also planned, and the rollout schedule will be announced in phases.

The energy research firm Noreva expects natural gas to pass $10 per million British thermal units at some delivery hubs in the coming years, against a Henry Hub price now just under three dollars. Meta has committed to a 7.5-gigawatt plant in Louisiana and Amazon to a 7.6-gigawatt plant in Texas, with Microsoft and Google building gigawatt-scale plants of their own. Chief executive Peter Gardett argues those buyers are carrying unusual price risk, which would surface either in token prices or on the wider grid.