Apple begins charging for artificial intelligence

Tim Cook approved what everyone was waiting for: Apple will sell expanded quotas for Apple Intelligence queries and the refreshed Siri through iCloud+ subscriptions. The basic tier will stay free with daily limits, but anyone who wants to work seriously with the new tools will have to upgrade to one of the pricier plans that will be added soon. The updated Siri is expected to arrive this fall as part of the next iOS release, after which it will be clear whether users are willing to pay for what has been free in the beta version.

google launches experiment to curb hallucinations

Google introduced Science One, an experimental framework that writes scientific papers without fabricating sources. The architecture, called Chain-of-Evidence, links each claim in the text to code that actually runs or to an existing artifact, abandoning the model’s internal memory in favor of real-time access to scientific APIs. An automated checker named CoE Audit verifies that citations are genuine, the code works, and the metrics reproduce. In a test using five systems built on Gemini 3.1 Pro, 75 papers were generated and all 337 citations were found to be authentic; in 12 of 15 experiments the code matched the declared methodology exactly.

new generation of video: longer and more controllable

In the video arena two Chinese players are pushing the limits. MiniMax released H3, a model that generates clips up to 15 seconds in 2K with stereo sound and can loop up to 12 references in a single prompt—camera movement from one video, a character from an image, a soundtrack from an audio file. Pricing on the API is 13 cents per second in 2K (about 48 agorot) and 9 cents for 768P. The open release on ModelScope is slated for 3 August. ByteDance follows with Seedance 2.5, which produces up to 30 seconds in a single pass—double the previous version—and allows chaining segments into minute-long videos. The model accepts up to 30 images, 10 clips and 10 audio tracks in one request, with time-codes for precise tempo control. In China it is available on Jiemeng AI and Doubao Professional, and worldwide through Dreamina on CapCut.

audio-box drops to $200 from $5,000

ElevenLabs completes the picture with Character Casting: a tool that breaks text into characters, assigns each a voice from a library of 10,000 synthetic speakers, and lets users swap a character’s voice throughout a document with a single click. The system can recognize invented names and terms; defining a pronunciation once applies it to the entire text. The company estimates the cost of producing an audiobook falls from $5,000 to $200, and production time shrinks from three months to a few hours. Support covers more than 70 languages, and export is available directly to ElevenReader.