Moonshot AI Unveils Kimi K3: The World’s Largest Open-Source AI Model Hits the Market

Moonshot AI, a Beijing-based artificial intelligence startup backed by Alibaba, has released Kimi K3, a 2.8-trillion-parameter model that the company claims is the largest open-source AI model globally.

Kimi K3’s Technical Specifications and Performance

Kimi K3 features a 1-million-token context window, native visual understanding, and an always-on reasoning mode called “thinking mode.” The model’s architecture includes two innovations: Kimi Delta Attention, a hybrid linear attention mechanism, and Attention Residuals, which the company describes as a drop-in replacement for residual connections. These were previously published as open research on GitHub.

Benchmark results show Kimi K3 competing closely with proprietary models like Claude and GPT. On GDPval-AA v2, it scored 1,687, placing it third behind Claude Fable 5 Max (1,815) and GPT-5.6 Sol Max (1,747.8). On AA-Briefcase, a private agentic benchmark, it ranked second with 1,527, trailing only Fable 5 Max (1,587). The model also achieved a state-of-the-art score of 91.2 on BrowseComp, a benchmark for long-horizon information seeking.

Strategic Implications and Market Position

The release of Kimi K3 represents a strategic move for Moonshot AI, which had seen its market position decline following DeepSeek’s rise. The company’s decision to open-source the model aims to position itself as a central hub for the global open-source AI developer community. According to Reuters, open-sourcing allows companies to showcase technological capabilities and expand influence, countering U.S. efforts to limit Beijing’s tech progress.

Kimi K3’s pricing at $3 per million input tokens and $15 per million output tokens aligns it with mid-tier offerings from Western labs. A promotional rebate through August 12 offers up to 30% back for API credits of $1,000 or more, per Moonshot AI.

Technical Demonstrations and Future Vision

Moonshot AI demonstrated Kimi K3’s capabilities by tasking it with designing a physical chip to run a nano-scale version of itself. Over 48 hours, the model completed the chip’s design pipeline using open-source tools, resulting in a 4mm² chip capable of decoding 8,700 tokens per second. The company also highlighted its ability to reproduce complex astrophysical calculations in two hours, as reported by Xinhua.

Market Context and Competitive Landscape

Moonshot AI, founded in 2023 by Tsinghua University graduate Yang Zhilin, raised $1.5 billion by early 2026. However, its user base declined after DeepSeek’s R1 model disrupted the market. The company’s pivot to open-source models, starting with Kimi K2 in 2025, aimed to reclaim relevance, with K3 as the culmination of this effort.

While Chinese tech giants like Alibaba and Baidu have released open-source models, none have matched Kimi K3’s parameter count.

Implications for Enterprise AI and Global Development

Kimi K3’s performance suggests the open-source and proprietary AI gap has closed at the frontier. Enterprises now have an alternative to API-dependent models, though running a 2.8-trillion-parameter model requires significant GPU infrastructure. Moonshot AI’s Mooncake project, which won a FAST 2025 award, aims to make such inference more practical.

China’s AI ecosystem, once questioned after chip export restrictions, now produces models rivaling those with access to advanced hardware.

Looking Ahead

Kimi K3’s release forces a reevaluation of enterprise AI strategies. As open-source models match proprietary systems in performance, the focus shifts from “productivity copilot” to “autonomous technical workforce.” With full model weights to be released on July 27, the AI community awaits independent validation of its claims.