Trade and Till
Money

Chinese AI Models Lead Open Model Race with Financial Focus

Recent free releases from Chinese AI firms, including Moonshot and Tencent, offer small businesses high-performance models for financial analysis at lower cost than leading closed alternatives.

Recent free releases from Chinese AI firms, including Moonshot and Tencent, offer small businesses high-performance...

Chinese AI companies are releasing powerful, open-weight models designed for business, with a clear lead in financial applications. Moonshot AI's Kimi K3.1 Pro model scored 58.6 on a finance agent test, surpassing OpenAI's GPT-6 Astra's score of 45.0. This is the fourth major Chinese open-weight release in recent months, following models from Tencent, Meituan, and Alibaba. Tencent launched its Hy4 model on November 4 and plans to release its open weights soon, trained specifically on financial workspaces like fund research and securities analysis.

Recent free releases from Chinese AI companies show aggressive scaling of model size and capabilities for business applications. Moonshot AI published the complete weights for its Kimi K3.1 Pro model, a 2.8-trillion-parameter system with a 1-million-token context window, on November 4. In a blind evaluation by 213 financial professionals, Tencent's Hy4 ranked first among open models. Kimi K3.1 Pro is the largest Chinese open-weight model by parameter count, surpassing DeepSeek's V4 Pro which has 1.6 trillion parameters.

Financial models prove enterprise readiness

New models specifically designed for financial workflows demonstrate measurable performance gains in real-world business tasks. Tencent claims to have cut Hy4's hallucination rate from 12.95% to 7.2%. The model is priced at $0.90 per million input tokens and $2.70 per million output tokens. Moonshot's Kimi K3.1 Pro leads Claude Opus 5.5 on 16 of 37 individual benchmark tests, though it trails on the aggregate Artificial Analysis intelligence index.

ModelKey Financial Benchmark ScoreParameter CountContext Window (tokens)
Moonshot Kimi K3.1 Pro58.6 (Finance Agent Test)2.8 trillion1,048,576
OpenAI GPT-6 Astra45.0 (Finance Agent Test)Not specifiedNot specified
DeepSeek V4 ProNot specified1.6 trillionNot specified

Cost efficiency enables small business adoption

Pricing structures and hardware requirements make advanced AI accessible to smaller operations. Kimi K3.1 Pro is priced at roughly half the cost of Claude Opus 5.5 via API. Running it requires approximately eight Nvidia B300 GPUs and 2.3 terabytes of GPU memory. Other global releases also lower barriers: Nvidia's Nemotron 3.5 Transcribe speech recognition model runs on a single GPU, enabling in-house transcription of customer calls without cloud dependency.

Regulatory and licensing frameworks shape access

Licensing terms and compliance requirements affect how businesses can deploy these models. The Kimi K3.1-Pro License allows use, modification, fine-tuning, and redistribution but requires commercial API providers with over $20 million in annual revenue to sign a separate agreement. Moonshot stated it avoids releasing open weights that extend the absolute frontier of cyber capability and keeps offensive security skills behind proprietary models public for at least three months. Similar licensing is seen elsewhere; Mistral AI's Large 4 model ships under a modified MIT license requiring large cloud providers to sign a separate agreement.

Other specialized models are becoming available. Z.ai opened its GLM-Legal model to public API access on November 2, designed for contract review and cross-border compliance. Washington has tightened chip export controls over the past year, but Chinese labs continue to release frontier-scale weights for free. Businesses can begin evaluating model suitability for financial tasks using the open weights now available from Tencent and other Chinese providers.

Topics

#Money

Related coverage

More from Money