Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: t.co/7m7eEg6Y0B Tech report: t.co/yeu6cjpMCT Tech blog: t.co/YTfiMSNM1f

Jul 27, 2026, 3:14 PM UTC

Views: 14,623,733
Likes: 45,900
Replies: 1,531
Reposts: 7,199
Quotes: 2,338
Bookmarks: 14,111

Ximage

Releasing the model weights and technical report of Kimi K3.

The post announces the release of Kimi K3 model weights and its technical report. The attached report page highlights the model’s architecture, long context window, visual understanding, efficiency claims, and benchmark results.

Views: 14.6M
Likes: 45.9K
Replies: 1.5K
Reposts: 7.2K
Quotes: 2.3K
Bookmarks: 14.1K
Topics: AI Model ReleaseTechnical ReportBenchmark ResultsAI

Analysis

Content overview

Formats

Research AnnouncementTechnical Report PreviewBenchmark Showcase

Topics

AI Model ReleaseTechnical ReportBenchmark Results

Sign in to see the full analysis.

Full post text

Releasing the model weights and technical report of Kimi K3. Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params. Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: https://t.co/7m7eEg6Y0B Tech report: https://t.co/yeu6cjpMCT Tech blog: https://t.co/YTfiMSNM1f

Tweet text hook

As written

Releasing the model weights and technical report of Kimi K3.

Tweet text structureThe Product Release Brief

From the post

01
The Announcement
Releasing the model weights and technical report of Kimi K3.
02
The Capability Case
Kimi K3 is our most capable model: a 2.8T MoE model with native visual understanding and a 1M-token context window. New model architecture: 2.5x the intelligence per unit of compute, not just more params.
03
The Expanded Access
Alongside Kimi K3, we're opening up more of the stack behind it — high-performance attention kernels, MoE communication library, and infrastructure for running agent environments at scale. Model weights: https://t.co/7m7eEg6Y0B Tech report: https://t.co/yeu6cjpMCT Tech blog: https://t.co/YTfiMSNM1f

Image and caption

Caption

Reusable caption template.

How the image and caption work togetherThe visual makes the announcement feel credible by showing the source document and measurable evidence. The caption gives the release context, summarizes the value, and directs readers to the resources. Together they pair proof with clear access points.

What stops the swipeA major release claim creates curiosity, while specific scale and performance details promise technical value. The visible report and benchmark charts reduce skepticism and invite researchers, builders, and technically minded readers to inspect the evidence.

Retention notes