Advertise
AI

Six AI model releases land, led by Kling 4.0 video model

Kling AI's Kling 4.0 generates clips up to 30 seconds, alongside new models from Fish Audio, Skywork, PrismML, kyutai and fal.

Six AI model and product releases were announced on October 15, 2026, according to aitimes.com, spanning video generation, speech, game worlds, small language models and robotics.

Kling AI released Kling 4.0, a video generation model that produces clips up to 30 seconds long and lets users specify multiple reference materials and key frames to keep people and scenes consistent. It is available on the redesigned Kling AI website and supports integration with external tools via MCP and a CLI.

Fish Audio unveiled a preview of Drama 3, a speech generation model that adjusts emotion and speaking style with natural-language instructions, inserts laughter, sighs and pauses, and applies whispering or emphasis to specific words. Drama 3 can generate multi-speaker dialogue in one pass and is available through a web service and API.

Skywork introduced Matrix-Game 4.0, which combines a game engine handling character positions and physics with an AI video model that converts simple 3D output into detailed real-time game visuals. It includes a coding agent that turns user requests into new game rules and is designed to run on a single consumer GPU.

PrismML released 2B, 4B and 9B parameter versions of Bonsai 2, ternary models based on Qwen3.5 with sizes of 0.43GB, 0.95GB and 2GB respectively, supporting reasoning, image understanding and tool calling. The weights are published under the Apache 2.0 license, and the 9B model generates up to 34 tokens per second on an iPhone 18 Pro Max.

kyutai released REKI, a 2.4B-parameter vision-language-action model pretrained only on public data that remembers recent camera scenes with little computation and roughly quarters redundant repeated robot motions in its own evaluation; the model weights plus training and evaluation code were published. fal also released fal Worlds, which lets users explore AI-generated virtual worlds in real time as the screen changes in response to their input.

Quick answers

How long are the clips Kling 4.0 can generate?

Kling 4.0 produces clips up to 30 seconds long and lets users specify multiple reference materials and key frames to keep people and scenes consistent.

What sizes does Bonsai 2 come in?

PrismML released 2B, 4B and 9B parameter versions of Bonsai 2, at 0.43GB, 0.95GB and 2GB respectively.

What can Drama 3 do?

Fish Audio's Drama 3 adjusts emotion and speaking style with natural-language instructions, inserts laughter, sighs and pauses, and can generate multi-speaker dialogue in one pass.

Source