Moonshot AI launches Kimi K3, a 2.8-trillion-parameter multimodal model with a one-million-token context window
The company claims Kimi K3 rivals GPT-5.6 and Claude Fable 5 across coding, automation and reasoning tasks
New subscriptions have been temporarily suspended as demand exceeded available GPU capacity
What Is Kimi K3? China's New AI Model With 2.8 Trillion Parameters That Rivalled GPT-5.6
Moonshot AI has introduced Kimi K3, a 2.8-trillion-parameter multimodal AI model with a one-million-token context window, claiming competitive performance against leading frontier models. The company temporarily halted new subscriptions after demand pushed its computing infrastructure close to capacity
What Is Kimi K3? China's New AI Model With 2.8 Trillion Parameters That Rivalled GPT-5.6
Summary
Advertisement
Chinese AI start-up Moonshot AI has launched Kimi K3, a 2.8-trillion-parameter model with native vision capabilities and a one-million-token context window, positioning it as a direct rival to Anthropic's Claude Fable 5 and OpenAI's GPT-5.6.
Within 48 hours, demand was so high that the company suspended new subscriptions to protect existing users.
Moonshot says K3 is designed for coding, scientific research, financial analysis, long-form knowledge work, image understanding, video editing and AI agents. The company claims the model has already completed complex engineering tasks autonomously, including optimising GPU kernels, building a Triton-like GPU compiler from scratch, designing a chip for running smaller AI models and creating a browser-based 3D game.
Moonshot also says K3 can independently review dozens of research papers, generate thousands of lines of code and build interactive dashboards for scientific analysis.
On benchmarks, Moonshot claims K3 outperforms GPT-5.5, Claude Opus 4.8 and GLM-5.2 across software engineering and automation tests, while remaining competitive with Claude Fable 5 and GPT-5.6Sol on several coding benchmarks.
Advertisement
| Benchmark | Kimi K3 | Claude Fable 5 | GPT 5.6 Sol | GPT 5.5 |
| DeepSWE | 67.5 | 70.0 | 73.0 | 67.0 |
| Program Bench | 77.8 | 76.8 | 77.6 | 70.8 |
| Terminal Bench 2.1 | 88.3 | 84.6 | 88.8 | 83.4 |
| FrontierSWE | 81.2 | 86.6 | 71.3 | 64.9 |
| SWE Marathon | 42.0 | 35.0 | 39.0 | 14.0 |
| BrowseComp | 91.2 | 88.0 | 90.4 | 84.4 |
| Automation Bench | 30.8 | 29.1 | 29.7 | 22.7 |
| GPQA Diamond | 93.5 | 92.6 | 94.1 | 93.5 |
| MMMU-Pro | 81.6 | 81.2 | 83.0 | 81.2 |
| OmniDocBench | 91.1 | 89.8 | 85.8 | 89.4 |
Subscriptions Paused
Although the reception was sharper than Moonshot had planned for, the company said in a post on X that it was halting new subscriptions temporarily to manage compute pressure.
"Kimi K3 has received far more love than we expected, and our GPUs are feeling it.
Over the past 48 hours, demand has pushed close to the limits of our current capacity. To protect the experience of existing subscribers, we're temporarily pausing new subscriptions and prioritising compute for current members. Existing subscribed users are not affected," Moonshot wrote.
The company also announced a structural change to its subscription model as it scales up. "We're adding capacity as fast as we can and will reopen new subscription spots in batches. Going forward, we'll also split membership into two more focused plans: Kimi Membership for Kimi Web, App, and Work, and Kimi Code Membership for coding workflows. This will help us match compute more precisely and keep the experience stable," it added.
Advertisement