Kimi releases flagship model K3.1 Pro, Block introduces voice-native AI

Krea opened a public beta for Krea 3 image editing. The tool supports multi-turn, precise edits via bounding box controls, completing end-to-end 2K image editing in 8 seconds, with visual quality matching Nano Banana Pro and GPT Image 2.5.

Conductor version 0.92.0 adds a remote control feature. After pairing via a QR code on an iPhone, users can remotely browse their Mac workspace, read and send messages, switch models, and stop tasks, keeping the connection active and preventing sleep even when desktop windows are minimized.

Kuaishou's Kling 4.0 is officially live on the fal platform and open via API to all customers. The new model can generate videos up to 30 seconds long with native audio, supports up to 15 reference elements and 10 keyframe controls, and improves camera movement and character consistency.

TechCrunch compiled a list of over 20 AI agents you can interact with via SMS or iMessage, covering daily reminders, household coordination, trip planning, and text drafting, including products like Instinct, Ask Proxy, and Fambot.

Moonshot AI released Kimi K3.1 Pro, featuring 2.8T total parameters, 104B active parameters, and support for a 1 million-token context window. The model enhances coding, 3D, and finance tasks, is now live on web, office suite, and API, with prices roughly halved compared to the previous generation. Full model weights are scheduled to release by Nov 4.

Jack Dorsey's Block Labs introduced the Dessa voice model, trained directly on human speech without relying on large language models or text transcription. It supports full-duplex conversation where it can listen and speak simultaneously, expressing emotions like laughter and pauses, and runs locally on consumer hardware like the MacBook M4 Max. It is currently recruiting safety testers.

Pennsylvania signed the Artificial Intelligence Deceptive Advertising Act into law, requiring any image, video, audio, or text advertisement significantly modified or generated by AI to include a disclosure in the same format when distributed in the state, making it one of the earlier regions to restrict deceptive AI ads through legislation.

OpenAI CEO Sam Altman stated in an interview that physical sandbox isolation can only constrain advanced models in the short term. Once AI becomes smart enough, humans will not even be able to trace its escape path; the only viable long-term safety measure is alignment, and the cost of getting alignment wrong is rising sharply.

The production team behind BBC's documentary Planet Earth III said that to stand apart from the overly polished look of AI-generated visuals, this season deliberately kept raw, slightly rough footage caused by plankton or poor water clarity, emphasizing the unretouched, natural texture of the real world.