Skip to main content

Kimi K3 Explained — What Moonshot Launched, and Where It Fits a Video Workflow

TL;DR — On 2026-07-16 Moonshot released Kimi K3, an open-weight MoE at about 2.8T total parameters (896 experts, 16 active) with a 1M-token context and native vision. Full weights followed on 2026-07-27. K3.1 Pro (2026-10-24) is the post-trained flagship at about 2.8T total / 104B active, and it takes text, image, and video input. K3 is $3/$15 per 1M tokens and K3.1 Pro is $2/$10 on OpenRouter. Both are in BibiGPT's model list — paste the link first, then let the model write.

2026-07-16 · K3.1 Pro 2026-10-24 2.8T MoE · 1M context Live demo below

Paste the video. Then pick Kimi K3.

Paste a long-form video or podcast URL. BibiGPT returns chapters and a transcript. Choose Kimi K3 or K3.1 Pro in the model list for the summary or script.

Add BibiGPT as a preferred source on Google See more BibiGPT in Top Stories and AI answers.

Key facts (90-second read)

Kimi K3 shipped on 2026-07-16 and its weights followed on 2026-07-27. K3.1 Pro shipped on 2026-10-24 with text, image, and video input at $2/$10 per 1M tokens, below K3's $3/$15. Both are in BibiGPT's model list. The workflow that makes sense: paste the link, let BibiGPT produce chapters and a timestamped transcript, then choose Kimi for the writing.

Features

What Kimi K3 is

Moonshot's flagship open-weight family, released on 2026-07-16 and expanded on 2026-10-24. Big on context and vision, and it reads a video directly in the newer version.

2.8T parameters, 16 experts active

A sparse mixture-of-experts model: 896 experts with 16 active per token, about 2.8T total parameters. Moonshot shipped it in two versions, K3 Max and K3 Swarm Max, first inside Kimi Code and the Kimi app.

1M-token context, native vision

Native context is 1M tokens, with vision built in rather than bolted on. That is enough for a book, a long transcript, or a full course in one call.

Open weights from 2026-07-27

Moonshot promised the full weights and released them on 2026-07-27 under a permissive, MIT-like licence. You can self-host or use it through a gateway — no vendor lock on the model itself.

K3.1 Pro takes video input

The 2026-10-24 refresh is a post-trained K3 at about 2.8T total / 104B active, still 1M context, with text, image, and video input. Full weights were planned for 2026-11-04.

Where Kimi K3 fits a video or podcast workflow

Kimi is also a direct competitor: it summarizes links, videos, and podcasts on its own. So the honest question is which half of the job you want it to do.

Price: $3 / $15, and $2 / $10 for K3.1 Pro

Per 1M tokens on OpenRouter, Kimi K3 is $3 input and $15 output. K3.1 Pro is $2 / $10 — cheaper than the model it post-trains, which is unusual and worth checking before you build a budget around K3.

Both are in BibiGPT's model list

Paste a YouTube, Bilibili, or podcast URL and BibiGPT returns chapters, a timestamped transcript, and a summary. Then pick Kimi K3 or Kimi K3.1 Pro for the longer writing.

A transcript first is still the reliable path

Handing a model a clean transcript costs far fewer tokens than handing it a 40-minute mp4, and you can check the quote against a timestamp. Use video input for short clips where you need the frames.

Kimi as the competitor to compare against

If you already use Kimi for link summaries, the difference is the output shape: BibiGPT keeps timestamps, chapters, and a mind map you can search, and lets you switch models per task.

5 things to know about Kimi K3

From Moonshot's release posts and the OpenRouter model cards.

  1. 1

    2.8T parameters, 16 experts active

    A sparse MoE with 896 experts, 16 active per token. Moonshot shipped two variants at launch: K3 Max and K3 Swarm Max, first in Kimi Code and the Kimi app.

  2. 2

    1M-token context and native vision

    Not an add-on: the 1M-token window and vision are part of the model. One call can hold a full book, a course, or an entire podcast season transcript.

  3. 3

    Open weights on 2026-07-27

    Moonshot released the full weights under a permissive, MIT-like licence. Self-host it or go through a gateway — the model is not locked to one vendor.

  4. 4

    K3.1 Pro adds video input

    The 2026-10-24 refresh is a post-trained K3 at about 2.8T total / 104B active, still 1M context, and it accepts text, image, and video. Full weights were planned for 2026-11-04.

  5. 5

    K3.1 Pro costs less than K3

    On OpenRouter, K3 is $3/$15 per 1M tokens and K3.1 Pro is $2/$10. A newer model at a lower price is unusual — check gateway pricing before you build a budget.

3 ways to use Kimi K3 with video and audio

The link becomes text first. Kimi writes what comes next.

A long lecture becomes structured notes

Paste the course URL. BibiGPT returns chapters and a transcript. Choose Kimi K3 to write the summary and notes per chapter.

A podcast becomes an article or script

The same transcript can feed a newsletter, a video script, or a lesson plan. Kimi's 1M context holds the whole episode without trimming.

You are weighing Kimi against other models

Kimi K3.1 Pro takes video, as do the Gemini 4 models. The GPT-6 family takes text, images, and files. Every family has its own page here.

Related BibiGPT pages

The summarizer handles the video. These pages cover the other model launches.

Sources

Launch facts come from Moonshot's posts and the OpenRouter model cards.

What is Kimi K3?

What is Kimi K3?

Moonshot AI's open-weight flagship, released on 2026-07-16. A sparse mixture-of-experts with about 2.8T total parameters (896 experts, 16 active), a 1M-token native context, and native vision. The full weights came out on 2026-07-27 under a permissive licence.

Loved by creators, students & researchers

Why people use BibiGPT to turn videos into text every day.

Trusted by 50,000+ users worldwide

★★★★★

“I paste a link and get clean captions in seconds — it saves me hours of retyping every single week.”

Maya R.

Content Creator · Repurposes short videos

★★★★★

“Exporting the transcript lets me review new words at my own pace instead of pausing the video constantly.”

Daniel K.

Language Learner · Studies with real videos

★★★★★

“Accurate, timestamped text I can quote directly. It has quietly become part of my daily workflow.”

Priya S.

Researcher · Cites public talks

Frequently Asked Questions

Ask us anything!

Popular guides

Get the transcript, then let Kimi K3 write.

Paste a YouTube, lecture, or podcast URL into BibiGPT. You get chapters, a transcript, and a mind map. Choose Kimi K3 or K3.1 Pro in the model list for the summary or the longer piece.