7/17/2026, 1:01:44 PM · foundation-models

Moonshot AI Releases Kimi K3, the World's Largest Open-Weight Model at 2.8 Trillion Parameters

The Beijing-based lab backed by Alibaba launched Kimi K3—a 2.8-trillion-parameter Mixture-of-Experts large language model—days before the 2026 World Artificial Intelligence Conference in Shanghai, claiming the largest open-weight release in AI history.

Overview

<cite index="1-14">Moonshot AI, the Beijing-based artificial intelligence startup backed by Alibaba, released Kimi K3—a 2.8-trillion-parameter model that the company says is now the largest open-source AI model in the world, and one that benchmarks show performs neck-and-neck with the most powerful proprietary systems from Anthropic and OpenAI.</cite>

<cite index="7-14,7-15">Kimi K3 went live on July 16, 2026, rolling out first on Kimi Code and inside the Kimi app, with two variants surfaced at launch—K3 Max for chat and agent tasks, and K3 Swarm Max for large-scale parallel processing.</cite> <cite index="1-17">Full model weights are scheduled to be released on July 27, according to details shared by researchers who reviewed the company's technical documentation.</cite>

Architecture and Capabilities

<cite index="2-7,2-8,2-9">Kimi K3 is a sparse Mixture-of-Experts (MoE) model built on two architectural updates—Kimi Delta Attention (KDA) and Attention Residuals (AttnRes)—both of which change how information flows across sequence length and model depth.</cite> <cite index="2-19">K3 uses Stable LatentMoE, effectively activating 16 of 896 experts.</cite>

<cite index="1-19">The model features a 1-million-token context window, native visual understanding capabilities, and an always-on reasoning mode that the company calls "thinking mode."</cite> <cite index="2-17">Moonshot states AttnRes delivers roughly 25% higher training efficiency at under 2% additional cost.</cite> Together, these design choices yield <cite index="2-26">roughly 2.5x better overall scaling efficiency than Kimi K2.</cite>

<cite index="9-11,9-12">In a blog post, Moonshot explained that Kimi K3's primary use case is focused on long-running autonomous software development tasks, designed to analyze large codebases, coordinate programming tools, and perform multistep tasks to achieve an end goal.</cite> <cite index="9-15,9-16">Moonshot said this is an example of a "vision-in-the-loop" system suited for game development, user interface design, and computer-aided design, with a demonstration showing a 3D open-world game that Kimi K3 reportedly built entirely in a web browser.</cite>

Competitive Position

<cite index="1-7">The company's own timeline chart of open-source frontier model scale positions K3 as a dramatic outlier, towering above competitors like DeepSeek (1.6T), Xiaomi (1.02T), and Alibaba (397B).</cite>

<cite index="6-9">On benchmarks, Moonshot says K3 delivers frontier-level performance and that among the models it tested, its overall intelligence ranks second only to Claude Fable 5 and GPT-5.6 Sol.</cite> <cite index="1-11">Its Mooncake project, which won the Best Paper award at FAST 2025, pioneered KV-cache-centric disaggregated serving for large language models—an architecture designed specifically to make inference at extreme scale more practical and cost-efficient.</cite>

Pricing and Availability

<cite index="6-13">Kimi K3 costs $0.30 per million tokens for cached input, $3 per million tokens for input that misses the cache, and $15 per million tokens for output, with the full 1,048,576-token context window included at that rate.</cite> <cite index="6-14">That output price sits above what Moonshot charged for K2.7 Code, reflecting the jump in scale, though it remains well under what OpenAI and Anthropic charge for their top-tier models.</cite> <cite index="1-22">On the application programming interface (API) side, Kimi K3 is compatible with the OpenAI software development kit (SDK), lowering the integration barrier for developers already building on OpenAI or Anthropic toolchains.</cite>

Strategic and Market Context

<cite index="1-15">The release, timed to land just ahead of the 2026 World Artificial Intelligence Conference (WAIC) in Shanghai, is a dramatic escalation in the global AI arms race and a watershed moment for the open-source AI movement.</cite> <cite index="14-15">The 2026 WAIC is being held in Shanghai from July 17 to 20.</cite>

<cite index="26-5">Moonshot AI was founded in 2023 by Yang Zhilin, a former Meta AI and Google Brain researcher.</cite> <cite index="22-1,22-2,22-3">The company raised $500 million in a Series C round boosting its valuation to $4.3 billion, led by IDG Capital with participation from Alibaba Group and Tencent Holdings.</cite> <cite index="26-1">By early 2026, that figure had more than doubled to $10 billion following a $700 million raise.</cite> Most recently, <cite index="25-3">Moonshot AI's latest funding round was $2 billion, closed on May 7, 2026.</cite>

<cite index="26-8">Moonshot's annual recurring revenue (ARR) topped $200 million in April, driven by rapid growth in paid subscriptions and API usage.</cite>

<cite index="10-10,10-11">In February 2026, Anthropic accused Chinese AI labs including Moonshot of "industrial-scale campaigns" to illicitly extract Claude's capabilities to improve their own models, with the leading US AI firm stressing the attacks were growing in intensity and sophistication.</cite> Moonshot has not publicly responded to those allegations.

Cross-references

Sources

  1. [1]
    China’s Moonshot AI releases Kimi K3, the largest open-source model ever, rivaling top U.S. systems | VentureBeat
  2. [2]
    Moonshot AI Releases Kimi K3: A 2.8 Trillion Parameter Open MoE Model With Kimi Delta Attention and 1M Context - MarkTechPost
  3. [3]
    Moonshot’s Kimi K3 pushes Chinese AI into Fable-level territory | Fortune
  4. [4]
    Moonshot's upcoming Kimi 3 is expected to close the gap with Anthropic's Opus 4.8 | TechCrunch
  5. [5]
    Moonshot AI to Launch Kimi K3, China's Largest AI Model | Phemex News
  6. [6]
    Moonshot's Kimi K3 To Be Largest Open Model With 2.8 Trillion Parameters, Has 1M Context Window
  7. [7]
    What Is Kimi K3? Moonshot's 2.8T, 1M-Context Flagship
  8. [8]
    China’s Moonshot Unveils Kimi K3, A 2.8-Trillion-Parameter AI Model - Finimize
  9. [9]
    China's Moonshot throws down the gauntlet with Kimi K3, the world's largest open-weights model - SiliconANGLE
  10. [10]
    China: Moonshot Release 2.8tn Parameter AI Rivalling the US
  11. [11]
    2025 World AI Conference to be held in Shanghai
  12. [12]
    2025 World AI Conference
  13. [13]
    WAIC 2025 – Artificial Intelligence International Institute (AIII)
  14. [14]
    Shanghai to host record-breaking World AI Conference from July 17 to 20
  15. [15]
    Attendee guide: 2025 World AI Conference_Latest
  16. [16]
    World Artificial Intelligence Conference (WAIC) 2025
  17. [17]
    Special Edition: World AI Conference Recap - AI Safety in China
  18. [18]
    WAIC-AIII – Artificial Intelligence International Institute (AIII)
  19. [19]
    世界人工智能大会
  20. [20]
    WAIC 2025 to showcase global AI innovation in Shanghai_Latest
  21. [21]
    Alibaba-backed startup Moonshot AI's valuation is up $500 million, sources say, after its rivals IPO in Hong Kong
  22. [22]
    Moonshot AI Raises $500M Series C at $4.3B Valuation
  23. [23]
    Moonshot AI Raises $500M Series C Funding to Boost AI Infrastructure
  24. [24]
    China AI Startup Moonshot Seeks $10 Billion Value in New Funding
  25. [25]
    Moonshot AI - 2026 Funding Rounds & List of Investors - Tracxn
  26. [26]
    China's Moonshot AI raises $2B at $20B valuation as demand for open source AI skyrockets | TechCrunch
  27. [27]
    Alibaba-Backed Moonshot AI Nears $4.8B Valuation in New Funding Round
  28. [28]
    IDG Leads $500M Series C for Moonshot AI, Oversubscribed by Alibaba & Tencent · TechNode
  29. [29]
    How Much Did Moonshot AI Raise? Funding & Key Investors | Clay
  30. [30]
    Moonshot AI - Wikipedia
Moonshot AI Releases Kimi K3, the World's Largest Open-Weight Model at 2.8 Trillion Parameters · AIDB