● Covered by 2 sources · 7 reportsMedium impact4 neutral
Kimi K3: New 2.8 Trillion-Parameter Open AI Model with Vision Capabilities Released
Key points
- Kimi K3 is a 2.8 trillion-parameter AI model.
- Model has native vision capabilities and a long context window.
- Performs well against proprietary models like Fable 5.
- Faces higher costs and longer task times than some models.
- Full model weights release planned for July 2026.
- Kimi K3 is built on Kimi Delta Attention and Attention Residuals.
- Kimi K3 has a 1-million-token context window.
- Kimi K3 is the world's first open 3T-class model.
- Kimi K3 is available on Kimi.com, Kimi Work, Kimi Code, and the Kimi API.
- Kimi K3 uses max thinking effort by default.
- Full model weights will be released by July 27, 2026.
- Kimi K3 showed comparable performance to Fable 5 across 1,030 tasks.
- Kimi K3 is selected for 72-96% of tasks in oracle routing.
- Kimi K3 is from Moonshot AI.
- Kimi K3 scores 1543 on the AA-Briefcase benchmark.
- Kimi K3 scores 57 on the Artificial Analysis Intelligence Index.
- Kimi K3's AA-Briefcase score is a +727 improvement over Kimi K2.6.
- UK AISI and CAISI conducted a joint evaluation of Kimi K3's cyber capabilities.
- Kimi K3 was released on July 16, 2026.
- The evaluation used the ExploitBench benchmark.
Introduction of Kimi K3
Kimi has unveiled Kimi K3, a new 2.8 trillion-parameter AI model, marking a notable advancement in open-source AI. It boasts native vision capabilities and a 1-million-token context window, designed for extensive coding, reasoning, and knowledge work tasks.
Performance and Comparison
Kimi K3 has demonstrated significant performance, especially in symbolic math and development tasks, achieving competitive scores against proprietary models like Fable 5. It scores 1543 on the AA-Briefcase benchmark, second only to Fable 5, and offers a potential cost-effective alternative for certain applications.
Challenges and Availability
Despite its strengths, Kimi K3 faces challenges, such as a significant cost of $10.57 per task and nearly an hour runtime per task, which may hinder widespread adoption. The model is currently available on Kimi platforms with full model weights to be released by July 2026.
Why Kimi K3 Matters
Kimi K3 represents a significant leap in open-source AI models, providing new options for developers and researchers. Its release could drive innovation in AI applications, despite its cost and performance trade-offs compared to other models.
Updates
- Kimi K3 is built on Kimi Delta Attention and Attention Residuals.
- Kimi K3 has a 1-million-token context window.
- Kimi K3 is the world's first open 3T-class model.
- Kimi K3 is available on Kimi.com, Kimi Work, Kimi Code, and the Kimi API.
- Kimi K3 uses max thinking effort by default.
- Full model weights will be released by July 27, 2026.
- Kimi K3 showed comparable performance to Fable 5 across 1,030 tasks.
- Kimi K3 is selected for 72-96% of tasks in oracle routing.
- Kimi K3 is from Moonshot AI.
- Kimi K3 scores 1543 on the AA-Briefcase benchmark.
- Kimi K3 scores 57 on the Artificial Analysis Intelligence Index.
- Kimi K3's AA-Briefcase score is a +727 improvement over Kimi K2.6.
- UK AISI and CAISI conducted a joint evaluation of Kimi K3's cyber capabilities.
- Kimi K3 was released on July 16, 2026.
- The evaluation used the ExploitBench benchmark.
✨ This summary was generated by AI from the outlets' reporting listed below. It is not independently verified and may contain errors — check the original sources. How BrevFeed works →
The daily brief
One email each morning: the day's tech stories, clustered across outlets and summarized. No account needed.
One email a day. Unsubscribe in one click, any time.
Today's brief
Spend a few minutes, get the whole day. Every topic's top stories in one hands-free rundown — listen, watch, or read the transcript.
▶ Play today's brief
New every morning, and the back catalogue is archived by date.
Primary sources
GitHub newrelic/newrelic-browser-agent arXiv 2605.14153 GitHub aws-samples/sagemaker-genai-hosting-examples GitHub awslabs/ai-on-eks-charts GitHub awslabs/ai-on-eksHow outlets covered it
Moonshot AI released Kimi K3, a 2.8 trillion parameter Mixture of Experts (MoE) model, on July 27, 2026, making its weights publicly available for self-hosting. This guide details how to deploy Kimi K3 on AWS using Amazon SageMaker HyperPod or Amazon EKS, addressing the infrastructure and serving framework requirements for such a large model.
This post details how to deploy Moonshot AI's Kimi K3, a 2.8 trillion parameter Mixture of Experts (MoE) model, on AWS using Amazon SageMaker HyperPod and Amazon EKS. Kimi K3 is the first open-weight model to approach the 3 trillion parameter class, offering frontier-level intelligence with publicly available weights for self-hosting.
The Kimi K3 model, a 2.8 trillion-parameter open-weight model, has been released, featuring a scaled-up architecture from its predecessor, Kimi Linear. Key architectural changes include the introduction of LatentMoE, efficiency tweaks for inference, attention residuals, and the use of NoPE (No Positional Embeddings) throughout, alongside native multimodal support.
The UK Artificial Intelligence Security Institute (UK AISI) and the U.S. Center for AI Standards and Innovation (CAISI) conducted a preliminary evaluation of Moonshot AI’s Kimi K3 model, focusing on its cyber capabilities. The assessment, primarily using the ExploitBench benchmark, indicates Kimi K3's ability in exploit development, contributing to the understanding of advanced AI models' potential in cybersecurity tasks.
Kimi K3, a 2.8T parameter AI model from Moonshot AI, scores 1543 on the AA-Briefcase benchmark, second only to Fable 5. However, it exhibits a significant average cost of $10.57 per task and takes nearly an hour to complete each task, indicating potential challenges for widespread adoption despite its strong performance.
Kimi K3, an open model, showed comparable performance to Fable 5 across 1,030 tasks, excelling in specific domains. The results indicate that K3 may offer a cost-effective alternative for AI applications, particularly in symbolic math and development tooling.
Kimi has launched Kimi K3, a 2.8 trillion-parameter AI model with native vision capabilities and a long context window. This model aims to improve performance in coding and reasoning tasks, marking a significant advancement in open-source AI capabilities.