InclusionAI Releases Ling 3.0 Flash Fin Finance Model

Ling 3.0 Flash Fin is a finance-domain fine-tune of the Ling 3.0 Flash model developed by InclusionAI.

InclusionAI launched Ling 3.0 Flash Fin on August 28, 2026. The model is available via the Vercel AI Gateway and OpenRouter. It appears on BenchLM release trackers.

One month of free API access applies to finance professionals and developers on OpenRouter.

Background on Specialized AI Models from Chinese Labs

Chinese AI laboratories continue to develop specialized models for targeted applications. Tencent released the Hy4 model. Alibaba issued updates to the Qwen family of models.

Ant Group supports the InclusionAI laboratory in these efforts. The focus remains on domain adaptation for high value sectors such as finance.

Launch Details and Availability

The release of Ling 3.0 Flash Fin took place on August 28, 2026. Model weights will become open source in the subsequent week.

API gateways facilitate immediate use. The one month free period applies to qualified users. The model is listed with benchmarks on release trackers.

Technical Specifications of the Model

The model retains the Ling 3.0 Flash architecture. Total parameters stand at 124 billion. Active parameters number approximately 5.1 billion.

The context window extends to 262 thousand tokens. This allows for extensive document processing. The mixture of experts design contributes to efficiency.

Optimization occurred through several stages. Financial corpora pretraining formed the initial step. Domain post-training followed. Tool-use optimization completed the process.

1. Financial corpora pretraining
2. Domain post-training
3. Tool-use optimization for annual reports, financial workbooks, multi-document research, valuation modeling, and Excel tasks

Key technical specifications and optimization details for Ling 3.0 Flash Fin

Specification	Value
Total Parameters	124 billion
Active Parameters	5.1 billion
Context Window	262K tokens
Release Date	August 28, 2026
Optimization Methods	Financial corpora pretraining, domain post-training, tool-use optimization

Benchmark Performance

Testing covered multiple finance oriented benchmarks. FinFIRST represents a joint effort with CICC.

Other evaluations include FinSearchComp Verified. Finance Agent forms another test. APEX-Agents, SpreadsheetBench, and τ³-Banking complete the suite.

The Finance Agent v2 benchmark yielded a score of 59.8 percent. This result surpasses the 57.9 percent achieved by Gemini 3.5 Flash.

Practical Demonstrations

A demonstration involved a seven sheet Google 2026 Q2 financials file. The file included more than 5000 formulas.

The model performed 23 consecutive tool calls. These calls addressed token usage disclosures.

Market and Stakeholder Implications

Finance professionals obtain a tool tailored for annual report review. Valuation modeling benefits from the domain adaptation.

Spreadsheet tasks receive support through the tool use capabilities. Developers integrate the model via established API services.

The open weights promise expands access for research and customization. Stakeholders in financial services gain from the specialized capabilities.

Expert Reactions

Ling 3.0 Flash Fin is InclusionAI’s finance-enhanced MoE language model, combining 124 billion total parameters with approximately 5.1 billion active parametersVercel AI Gateway

The emphasis on practical financial applications draws attention from industry participants.

Outlook for Subsequent Releases

Open sourcing of the weights occurs the week after the August 28 launch. Further refinements may appear based on user feedback.

The model positions InclusionAI within the competitive landscape of specialized frontier models.