Moonshot AI, the Beijing-based artificial intelligence startup backed by Alibaba, released Kimi K3 on Thursday. The headline number is the point: it’s a 2.8-trillion-parameter model that the company says is now the largest open-source AI model in the world, with benchmarks that it claims run “neck-and-neck” with top proprietary systems from Anthropic and OpenAI.

The timing matters too. Moonshot is landing this release just ahead of the 2026 World Artificial Intelligence Conference in Shanghai, and it also gave a concrete schedule for the part that tends to move markets: full model weights are scheduled to be released on July 27, according to details shared by researchers who reviewed Moonshot’s technical documentation. If you want to test it immediately, you can sign up at kimi.com with a Google account or phone number, and start chatting without a credit card.

Zoom in on what Kimi K3 is actually built to do, and it gets harder to dismiss as a “bigger number” flex. Under the hood, it’s a frontier-class large language model with 2.8 trillion total parameters. Moonshot pairs that scale with a 1-million-token context window, native visual understanding capabilities, and an “always-on reasoning mode” the company calls “thinking mode.” The company also says the architecture is driven by two internally developed innovations: Kimi Delta Attention, a hybrid linear attention mechanism, and Attention Residuals, described as a drop-in replacement for residual connections that delivers consistent scaling gains. Moonshot previously published those techniques as open research on GitHub, which is a quiet but important signal: this is not a closed, black-box engineering story. It is an openly explainable path to scale.

Then there’s the “integration reality” layer. On the API side, Moonshot says Kimi K3 is compatible with the OpenAI SDK, which lowers the switching cost for developers already building on OpenAI or Anthropic-style toolchains. That matters operationally because teams rarely rebuild from scratch, especially when product roadmaps are already full. Pricing also looks designed to be hard to ignore: $3 per million input tokens and $15 per million output tokens, with cached input tokens dropping to $0.30 per million. Moonshot says that places it roughly in line with mid-tier offerings from Western labs, while also claiming performance near the top of the market. There’s even a promotional rebate running through August 12, offering up to 30 percent back in vouchers for API credits of $1,000 or more.

The performance claims are where the open-source narrative either holds or breaks. Moonshot points to benchmark results drawn from public leaderboard data and a private evaluation by Artificial Analysis. On GDPval-AA v2, which measures real-world tasks across 44 occupations and 9 major industries, Kimi K3 scored 1,687, landing third overall behind Claude Fable 5 Max (1,815) and GPT-5.6 Sol Max (1,747.8), and ahead of Claude Opus 4.8 (1,600). On AA-Briefcase, a private agentic benchmark from Artificial Analysis designed for long-horizon knowledge work, K3 climbed to second place with 1,527, beating GPT-5.6 Sol Max (1,495) and trailing only Fable 5 Max (1,587). And on BrowseComp, a benchmark for long-horizon, high-difficulty information seeking, K3 posted a state-of-the-art score of 91.2 out of 100.

Moonshot’s “how we did it” detail is particularly telling. The company says it achieved that BrowseComp result in a single-agent setup using its 1-million-token context window, without any context compression or additional context management techniques. In other words, the argument is not only that Kimi K3 is good, but that long context plus retrieval and raw capacity can beat elaborate multi-agent patchwork. One widely followed AI commentator put it on social media: “Open source is no longer lagging six months behind Western closed-source models. Read that again, and think about what it all means.”

Moonshot is also showcasing a proof-of-concept that points to a next phase beyond text generation. In a demonstration documented in the company’s technical materials, Kimi K3 was tasked with designing a physical chip to run a nano-scale version of itself. Over 48 hours of continuous autonomous agent operation, K3 independently completed the chip’s full construction pipeline, from architectural design through optimization and verification, using open-source electronic design automation tools. The result was a tiny but functional chip design, 4 square millimeters, hitting timing convergence at 100 MHz and decoding more than 8,700 tokens per second in simulation. Moonshot frames this as not a production chip, but a demonstration of long-range autonomous agent capabilities.

The company also highlighted computational astrophysics work, saying K3 reproduced the universal I-Love-Q relation, typically taking a senior researcher one to two weeks, in approximately two hours. Moonshot says it accomplished this by reading and cross-validating more than 20 papers and implementing a complete numerical pipeline along the way. Those examples are important because they suggest a strategic shift: the race is increasingly about sustained, multi-step technical work over hours, not just crisp answers in a single turn.

Finally, the backstory explains why this release feels like a comeback and a repositioning at the same time. Founded in 2023 by Yang Zhilin, a Tsinghua University graduate who previously conducted research at Google and Meta, Moonshot AI gained early traction in 2024 as users flocked to its Kimi platform for long-text analysis and AI search functions. By early 2026, it had raised roughly $1.5 billion across multiple rounds, with valuation rising from $2.5 billion to $4.3 billion, and the company reportedly seeking a new round at $5 billion. Then DeepSeek arrived. After DeepSeek’s low-cost R1 model launched in January 2025, Moonshot slid: Kimi went from third in monthly active users in China to seventh. Moonshot’s pivot toward open-source began with Kimi K2 in July 2025 and accelerated with K2.5 in January 2026.

Kimi K3 looks like the culmination of that plan, and the scale suggests deep preparation because training a 2.8-trillion-parameter model requires enormous computational resources and months of work. For executives, boards, and product leaders, the second-order implication is straightforward: if open-source models are reaching the top end on real benchmarks while staying compatible with mainstream developer tooling and competitive pricing, then the procurement conversation shifts from “can we access the model?” to “what will we build, and how fast can we adopt and differentiate?” In the global AI arms race, Kimi K3 is Moonshot pushing from the sidelines into the main event, with full weights still two weeks away on July 27.