1littlecoder · 2026-07-16 · notable
1littlecoder: '2.8 Trillion Parameters - Kimi K3 is here'
1littlecoder covers Moonshot's Kimi K3 launch on release day: the 2.8T Mixture-of-Experts architecture, 1M-token context, and how the new flagship compares to Kimi K2.7, GPT-5.5, and Claude Opus 4.8.
1littlecoder walks through Kimi K3 on release day — a 2.8T-parameter Mixture-of-Experts from Moonshot with a 1M-token context.
What is it?
The July 16 video from 1littlecoder covers Moonshot's Kimi K3 launch. K3 is the successor to Kimi K2.7, going from 1T to a 2.8T Mixture-of-Experts, extending the context window from 256K to 1M tokens, and adding native vision and always-on thinking mode.
How does it work?
1littlecoder demonstrates Kimi K3 through kimi.com and the Kimi Code coding agent, comparing its output on coding prompts and long-context work against Kimi K2.7 and Claude. The video also walks through the new Delta Attention hybrid architecture Moonshot describes in the quickstart docs and points at the OpenAI-compatible `kimi-k3` API endpoint at api.moonshot.ai/v1.
Why does it matter?
Kimi K3 is the first Chinese frontier model to cross 2 trillion parameters, and 1littlecoder is one of the fastest same-day explainers of new Chinese model drops for a Western developer audience. The video gives viewers a hands-on look before Moonshot publishes a model card or weights, which lets teams decide whether to route production traffic to K3 today.
Who is it for?
Developers deciding whether to test Kimi K3 against Claude Opus 4.8 and GPT-5.5 on real workloads.
Try it
https://www.youtube.com/watch?v=lIPs0fj2Bgc