Visual Intelligence News
Kimi K3 Arrives as a 2.8T-Parameter Multimodal Model
Moonshot AI released Kimi K3 on July 17 as a 2.8-trillion-parameter, natively multimodal model with a one-million-token context window. The model is available through Kimi products and its API now; Moonshot says full weights will follow by July 27.
What happened and why it matters
The release joins extreme model scale with native vision and long-horizon tool use, while its weights, technical report and independent evaluations are still pending.
Primary source
Primary reference: Moonshot AI: Kimi K3 Open Frontier Intelligence. Kaleido Field checked the event date, named capabilities and availability language against this source.
| Source date | July 17, 2026 |
|---|---|
| Checked by Kaleido Field | July 18, 2026, 09:05 CST |
| What this source supports | primary-source model release with benchmark and availability boundaries for Kimi K3 2.8 trillion parameter multimodal model native vision one million context |
| What it does not prove | It does not prove a universal product ranking, full regional availability, or performance on every visual intelligence task. |
What Moonshot released
Kimi K3 uses a mixture-of-experts design with 2.8 trillion total parameters and activates 16 of 896 experts. Moonshot lists native image and video understanding, a one-million-token context window and long-horizon coding and research as core functions.
The model is available through Kimi, Kimi Work, Kimi Code and the Kimi API. Moonshot says full model weights will be released by July 27 and that a technical report with further architecture and training detail will follow.
The visual capability
Moonshot says K3 can inspect screenshots while editing code, reason about 3D scenes, work from images and video, and revise visual output during game, CAD and frontend tasks. It also reports multimodal benchmark results averaged across repeated runs.
Those results come from Moonshot's release evaluation. Native vision is a verified product claim, but the quality of perception, video editing and visual-agent work still needs independent testing across real tasks.
Limits in the release itself
Moonshot says K3 still trails Claude Fable 5 and GPT-5.6 Sol overall and warns that incomplete thinking history can destabilize output. It also flags excessive proactiveness when the model meets ambiguous instructions.
The company has published pricing and availability, but the weights and full technical report are not yet public. Claims about coding, chip design and research speed should remain attributed to Moonshot until artifacts and independent reproductions are available.
Evidence boundary
This page reports a dated event from a named primary source. Company specifications and adoption statements remain attributed claims unless independent evidence is cited above.
FAQ
What is the practical answer?
Moonshot AI released Kimi K3 on July 17 as a 2.8-trillion-parameter, natively multimodal model with a one-million-token context window. The model is available through Kimi products and its API now; Moonshot says full weights will follow by July 27.
What source does this article use?
The primary source is Moonshot AI: Kimi K3 Open Frontier Intelligence. Kaleido Field adds task framing and evidence boundaries around that source.
Where should the user verify the answer?
Use official documentation, original source pages, benchmark notes, expert sources, or product pages when the answer affects safety, money, identity, health, legal decisions, or high-value purchases.