Moonshot AI releases Kimi K3, China's largest model at 2.8 trillion parameters
Beijing-based Moonshot AI released Kimi K3, China's largest AI model at 2.8 trillion parameters. The company claims the model consistently outperforms OpenAI's GPT 5.5 and Anthropic's Claude Opus 4.8 on benchmarks including coding and general agents, though it still trails the leading-edge GPT 5.6 Sol and Claude Fable 5 in overall performance.
Kimi K3 — Quick Specs
Moonshot AI releases Kimi K3, China's largest model at 2.8 trillion parameters
Beijing-based Moonshot AI released Kimi K3 on Friday, China's largest AI model to date with 2.8 trillion parameters. The company claims the model consistently outperforms OpenAI's GPT 5.5 and Anthropic's Claude Opus 4.8 on benchmarks including coding and general agents, though it still trails the leading-edge GPT 5.6 Sol and Claude Fable 5 in overall performance.
The model represents what Bank of America analysts called "step-change gains for flagship Chinese models" despite persistent hardware and compute capacity constraints in China due to U.S. export controls.
Performance claims
According to Moonshot AI, Kimi K3 beat Claude Opus 4.8 and GPT 5.5—models that sit just behind Anthropic and OpenAI's most advanced systems—on coding and general agent benchmarks. The company acknowledged that K3 still trails Anthropic's Claude Fable 5 and OpenAI's GPT 5.6 Sol on overall performance.
Specific benchmark scores and pricing details were not disclosed at the time of release.
Market impact
Chinese AI rivals' shares dropped sharply on news of the release. DeepSeek, which released a model in June, saw its stock fall 28% on Friday. MiniMax Group dropped 16%. Alibaba, which makes the Qwen series of models, fell 4% despite earlier gains from an Apple partnership announcement.
"K3 raises the capability ceiling for China AI models, shifting the burden of proof to other independent AI labs," said Bank of America analyst Alex Liu in a research note.
Company background
Founded in 2023, Moonshot AI is one of China's leading model builders. The company has focused on large-scale pre-training and architectural innovation to compete with U.S. labs despite hardware restrictions.
The release comes as Chinese AI models gain traction among Western companies, offering competitive performance at lower prices than advanced U.S. offerings. U.S. lawmakers are considering measures to curb adoption of Chinese AI models by American companies.
What this means
Kimi K3's release signals China's ability to build competitive large language models despite semiconductor export controls that limit access to advanced chips. The 2.8 trillion parameter count represents a significant scale achievement, though parameter count alone doesn't determine model quality. The market reaction—with rival Chinese AI companies losing significant value—suggests investors view Moonshot AI as a credible competitive threat in China's AI ecosystem. Without independent benchmark verification and pricing details, it remains unclear how K3 compares to U.S. models in real-world applications.
Related Articles
DeepSeek Releases V4-Flash-Vision-Exp, First Multimodal Model in V4 Family
DeepSeek has released DeepSeek-V4-Flash-Vision-Exp, its first experimental multimodal model in the V4 family, adding visual understanding to the V4-Flash architecture. The 305B-parameter model shows substantial gains on multimodal agent benchmarks while holding steady on text-only tasks.
Tencent Open-Sources Hy4 Preview: 770B-Parameter MoE Model with 1M-Token Context
Tencent's Hy Team has open-sourced Hy4 preview, a 770-billion-parameter Mixture-of-Experts model with 49 billion activated parameters and a 1-million-token context window. The model is available under Apache 2.0 alongside an FP8-quantized variant, with Tencent claiming it beats GLM 5.3 and Kimi K3 on internal engineering evaluations.
Tencent Releases Hy4 Preview: 770B-Parameter MoE Model with 1M Context for Coding Agents
Tencent has released Hy4 preview, a mixture-of-experts model with 770B total parameters and 49B active parameters, targeting coding agents and multi-step tool-use workflows. The model ships with a 1 million token context window and is priced at $0.834 per 1M input tokens and $2.501 per 1M output tokens.
Google Launches Gemini 3.5 Transcribe with 4.0% Word Error Rate Across 85 Languages
Google has released Gemini 3.5 Transcribe, a speech-to-text model that automatically detects 85 languages, removes filler words, and corrects misspoken phrases. The company claims a 4.0 percent word error rate for streaming audio and 70 percent lower latency than its predecessor, Chirp 3.
Comments
Loading...