News — 2026-08-16

A tech writer recently swapped careful prompt crafting for a ten‑minute stream of consciousness in ChatGPT’s voice mode. Speaking freely about vacations, TV choices, birthday gifts and more, the author let the AI sort the jumble afterward. ChatGPT identified a unifying theme—making everyday moments more meaningful—and offered practical suggestions. The experiment suggests that unedited speech can surface contradictions and hidden priorities that polished prompts often erase, giving the model a fuller picture of a user’s intent.Read more

OpenAI has added a new "Computer History" feature to ChatGPT for macOS users on its Pro, Business and Enterprise plans. The optional tool records clicks, keystrokes, app switches and other activity via macOS’s accessibility API, letting the AI summarize recent actions and suggest repeatable tasks. Users can pick which apps or sites to track, pause logging, and delete the history at any time. The feature is not yet available in the European Economic Area, Switzerland or the United Kingdom, and administrators can control opt‑in settings for those regions.Read more

DeepSeek announced a steep increase in its AI API pricing effective Aug. 16, as the company rolls out its V4 Pro and V4 Flash models. Peak‑hour rates for V4 Pro will jump to $3.96 per million output tokens, more than four times the current $0.87 price, while off‑peak rates will be $1.98. V4 Flash will cost $1.32 at peak and $0.66 off‑peak, up from $0.28. The move ends a promotional period that began in May and positions DeepSeek’s new rates still well below rivals like OpenAI and Moonshot.Read more

OpenAI and Anthropic have slashed the prices of their mid‑tier large‑language models, aiming to make U.S. offerings more competitive against a wave of lower‑cost Chinese AI services. OpenAI cut its GPT‑5.6 Luna model to $0.20 per million input tokens and $1.20 per million output tokens, while Anthropic introduced Opus 5 at $5 per million input and $25 per million output tokens, half the cost of its flagship Fable 5. The moves accompany a canceled price hike for Anthropic’s Sonnet 5 and signal a strategic push to defend the high‑end market while courting cost‑sensitive developers.Read more

Google announced Friday that users will soon be able to disable the visible watermark on images, videos and songs generated by its Gemini models. The change, which applies to Nano Banana, Omni and Lyria, leaves the invisible SynthID watermark and C2PA metadata intact for transparency. Settings will be added to Gemini, the Flow video editor and later to Search, with the feature rolling out in the next few days.Read more

OpenAI’s chief financial officer Sarah Friar told shareholders on Friday that the company’s enterprise segment now generates more revenue than its consumer‑facing ChatGPT service. The AI firm’s annualized run rate has climbed to $40 billion, roughly double the figure a year ago, and July revenue rose 20% month‑over‑month with business customers up 32%. Friar highlighted a shift among enterprise users toward cost‑per‑intelligence metrics and noted recent price cuts and efficiency gains in OpenAI’s model lineup. The announcement came amid executive departures and a pending SEC filing about a potential listing.Read more