Moonshot AI launches Kimi K3

2026-07-16 · Source: Constellation Research · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Emerging Technologies & Innovation · Depth: Advanced, quick

Summary

Moonshot AI launched its Kimi K3 model on July 16, 2026, a 2.8T parameter large language model positioned to rival Anthropic's Claude Opus 4.8 and OpenAI GPT-5.5. This open-weight model features a 1 million token context window and native vision capabilities. Kimi K3 is a scaled-up mixture-of-experts model, built on Kimi Delta Attention (KDA) and Attention Residuals (AttnRes) architectural updates, with optimized GPU kernels. The company also demonstrated a chip design based on the model as a proof of concept. Pricing is set at 30 cents per million cached input tokens, \$3 per million non-cached input tokens, and \$15 per million tokens for output. Full model weights are expected by July 27.

Key takeaway

For AI Scientists evaluating large language models, Kimi K3 presents a compelling open-weight option with frontier-level performance. Its 2.8T parameters, 1 million token context, and native vision capabilities, coupled with competitive pricing, warrant your immediate consideration for projects where cost-efficiency and advanced features are critical. Assess its benchmarks against your specific use cases.

Key insights

Kimi K3's release intensifies open model competition, potentially narrowing the performance gap with proprietary LLMs.

Principles

• Open models are rapidly closing the performance gap.
• Architectural updates can optimize GPU kernels.

Method

Kimi K3 employs Kimi Delta Attention (KDA) and Attention Residuals (AttnRes) for architectural updates and GPU kernel optimization.

In practice

• Utilize 1 million token context window.
• Integrate native vision capabilities.
• Consider scaled mixture-of-experts models.

Topics

• Moonshot AI
• Kimi K3
• Large Language Models
• Open-weight Models
• Mixture-of-Experts
• LLM Pricing
• Vision Capabilities

Best for: CTO, VP of Engineering/Data, AI Engineer, AI Scientist, Machine Learning Engineer, Director of AI/ML

Related on AIssential

• Moonshot’s Kimi K3 closes the frontier gap — Moonshot AI's Kimi K3 demonstrates open-source models can achieve frontier-level performance at significantly lower costs.
• Kimi K3: The Biggest Open Model Ever, Except You Can’t Have It Yet — Moonshot AI's Kimi K3 is a 2.8 trillion parameter model marketed as open, but its weights are not yet publicly available.
• Kimi K3 just beat FABLE. — Chinese open-source models, like Kimi K3, are now frontier-class competitors to proprietary LLMs.
• Moonshot’s Kimi K3 May Be More About Memory Than Compute — Moonshot's Kimi K3 model may prioritize memory over compute for performance.
• 10 Best Use Cases of Kimi K3 (The Model That Made US Stocks Fall Yesterday) — Kimi K3 is a powerful open-weight model challenging frontier LLMs on specific tasks despite its high cost.
• Kimi K3: Can Moonshot’s New Model Be Actually Used For Free? — Kimi K3 is a 2.8T parameter model with a 1M token context window, specialized for coding and autonomous agents.

See Counsel's argued verdicts on the open AI decisions leaders are weighing →

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Constellation Research.