0:00 / 3:18
Chapters
Sources
- 01. Moonshot AI Releases Kimi K3, a 2.8 Trillion Parameter Open-Weight Model open_in_new
- 02. Google Brings Gemini Omni Video Editing and Personal Avatars to Vids open_in_new
- 03. Google Flow Music Now Lets You Upload Your Own Beats and Instrumentals open_in_new
- 04. DoorDash Opens dd-cli Beta for AI Agents to Order Food Autonomously open_in_new
DAILY ROUNDUP
China's Kimi K3 Rivals Claude and GPT. Google Vids Gets AI Editing. DoorDash Gets a CLI.
calendar_today Date:
schedule Duration: 3:18
visibility 1 Views
Moonshot's Kimi K3 rivals Claude and GPT, Google Vids gets conversational AI editing, Flow Music takes your own beats, and DoorDash opens a CLI for agents.
- 01. Moonshot AI releases Kimi K3, a 2.8 trillion parameter open-weight model rivaling Claude and GPT-5.6 on coding benchmarks
- 02. Google brings Gemini Omni to Vids for conversational video editing plus personal AI avatars
- 03. Google Flow Music now lets you upload and remix your own beats and instrumentals
- 04. DoorDash opens a limited beta of dd-cli, letting AI agents autonomously order and pay for food
Moonshot AI has released Kimi K3, a 2.8 trillion parameter open-weight model the company calls the largest open-source AI model in the world. It uses a mixture-of-experts design, meaning only a fraction of those parameters activate per request, and it ships with a million-token context window and native visual understanding. Moonshot says K3 still trails Fable 5 and GPT-5.6 Sol on overall performance, but it outperformed every other model in the company's test suite, including Claude Opus 4.8, on coding and agentic benchmarks. Full weights are due out later this month, timed just ahead of Shanghai's World AI Conference.
Google has brought Gemini Omni into Vids, allowing users to edit video by simply describing the change they want, whether that's swapping a background, fixing lighting or cleaning up messy footage. The feature supports step-by-step edits, so each new instruction builds on the last rather than starting over. A personal avatar feature also lets users generate a digital version of themselves for videos they never actually filmed. Every clip carries an invisible SynthID watermark so it can be verified as AI-generated. It's available now for Workspace and Google AI Pro and Ultra subscribers.
Google's Flow Music now lets users upload their own beats and instrumentals rather than generating everything from scratch. Drop in a track or a rough idea and Flow Music analyses the style, key, tempo and instrumentation, then builds new sections to match. Users can shape the result with plain-text instructions, such as making the intro more energetic or stripping it back to instrumental only. Full songs run up to three minutes, and short loops can be extended to that length too, shifting the tool from pure generation towards remixing existing material.
DoorDash has opened a limited beta of dd-cli, a command-line tool that lets an AI agent search restaurants, compare deals, build a cart and check out entirely on its own. This isn't a sandboxed demo: it processes real payments and places real orders. Access is invite-only for now, requiring a waitlist application with a social link and stated use case. Anthropic has already demonstrated Claude completing a full DoorDash order autonomously from start to finish, suggesting agent-placed food orders will become increasingly common.
Meta Data
Company:
Chapters
Share
Actions
Sources
- 01. Moonshot AI Releases Kimi K3, a 2.8 Trillion Parameter Open-Weight Model open_in_new
- 02. Google Brings Gemini Omni Video Editing and Personal Avatars to Vids open_in_new
- 03. Google Flow Music Now Lets You Upload Your Own Beats and Instrumentals open_in_new
- 04. DoorDash Opens dd-cli Beta for AI Agents to Order Food Autonomously open_in_new
