The Week AI Moved Into Everyone's Hands

By Crypt0 · Oct 23, 2026

This week the frontier moved into everyone's hands. Kuaishou opened the Kling 4.0 video model to developers through a public API, Hi3D released a production tested 3D model as open source, Moonshot's Kimi K3.1 went from a code string to an official video reveal, and DeepSeek wrapped a funding round near $15 billion to build data centers and open models. The pattern repeats across all four stories. Powerful AI is becoming something you can download, plug into and ship with today.

The Kling 4.0 API service officially launched on October 22 and is opening to enterprise customers and developers worldwide. SelfyzAI announced the same week that it is among the first platforms to integrate the API, after joining the model's first batch internal beta ahead of release. Headline capabilities include native single takes of up to 30 seconds, up to 10 keyframes per generation for shot level control, Omni Reference with up to 15 combined reference inputs across images, video, voice and subjects, and native two channel stereo audio with tighter lip sync. The release builds on the full model that rolled out October 15, doubling Kling 3.0's 15 second ceiling and adding 4K output, a 21 to 9 ultrawide aspect ratio, and prompts up to 8,000 tokens. For creators, a 30 second take means more footage to cut from, and keyframe plus reference control keeps characters and products consistent across clips. The model layer is becoming the raw footage supply for an entire creator economy.

Beijing based Hi3D announced the open source release of Hi3D V1.0 on October 23, the image to 3D model previously called Sparc3D. The release puts source code on GitHub and model weights on Hugging Face, with free generation on the company's own site. The training set grew from about 500,000 to 2 million curated 3D assets, and the model extends progressive geometry reconstruction to an effective sparse grid resolution of 1536 cubed. Hi3D also cut credit consumption by more than 50 percent. The fine print matters as much as the release. Independent analysis notes the release ships under the Hi3D Open License Agreement Version 1.0, which sets its own rules. Products passing 200,000 monthly active users or $1,000,000 in monthly revenue must notify the licensor and apply for a separate written licence within 60 days, commercial integrations carry a 'Powered by Hi3D 1.0' branding line, and the code is inference only, with training code and the evaluation harness held back. The reference build wants Linux, an NVIDIA GPU with compute capability 8.0 or newer, and CUDA 12.0 to 12.5, with the top setting validated on a 96 GB H20, though a community MLX port landed the same day for Apple silicon. Open weights are a genuine building block, and reading the licence first is part of the build.

Moonshot's next model went from whispers to official this week. Kimi released a 24 second retro computer style video displaying Model Kimi K3.1 inside the Kimi Code terminal, with the model generating complex Three.js 3D scene code. Developers had already spotted the model's identifier string on the Moonshot API platform in late September, and K3.1 appeared in Kimi Code's model switching configuration on October 21. The Deploy Log's record for the week of October 19 to 25 lists Kimi K3.1 and MiniMax M3.1 as the two models first seen this week, and records Kimi K3.1 Pro as available on OpenRouter at $2 per million input tokens and $10 per million output tokens, with a one million token context window. That pricing puts a frontier coding assistant within reach of independent builders, which is exactly the crowd that moves fastest with new models.

The money story of the week belongs to DeepSeek. The Hangzhou lab is wrapping a funding round of about 100 billion yuan, roughly $15 billion, one of the largest in China's history. Reuters reported, citing four people with knowledge of the matter, that battery maker CATL and WeChat operator Tencent are among the biggest investors, with more than 15 investors in total and a pre money valuation of 500 billion yuan. The round follows a June raise of more than $7 billion at a $50 billion valuation, and the company is reported to be preparing for a potential STAR Market listing. TechCrunch noted the timing, two days after US federal civilian agencies were ordered to remove DeepSeek's technology from their systems over security concerns. The reported plan for the capital is computing infrastructure and open models for domestic and international markets. Capital is flowing toward open models at a scale the industry has rarely seen.

A question I keep hearing. What actually changes for an everyday builder this week? The short answer is that the download became the distribution. A video model you can call through an API, a 3D model whose weights you can self host, and a coding model at two dollars per million tokens all arrived within days of each other. A research lab or a corporate budget stopped being the price of admission. A weekend and a clear idea now do the job.

For readers who build, the signal is clear. The models are leaving the lab and landing in your hands, priced like software instead of science projects. Start with one of this week's releases, ship something small, and learn by doing. The frontier just became a toolkit, and the first movers will define what gets built.

Quick answers

What is this story about?

Why does this story matter?

Sources

New to crypto? Read the crypto glossary, browse frequent questions, read our story, or explore the story archive.