Ling 3.0 Flash Fin

Artificial Analysis published an independent evaluation of Ling-3.0-flash-Fin on September 16, 2026, scoring it 23 on the Artificial Analysis Intelligence Index and 24 on the Finance & Accounting Index, placing it on the Intelligence-vs-Active-Parameter Pareto Frontier. The text-only model was developed by Ant Group wi Key comparative findings: Ling-3.0-flash-Fin matches MiniMax-M2.7's Intelligence Index score of 23 while activating roughly half the active parameters (5.1B vs 10B), and ties Ling-3.0-flash-VL at 24 on the Finance & Accounting Index with higher business-knowledge accuracy (17% vs 11%) but also higher business-knowledge

Ant Group announced the open-sourcing of Ling-3.0-flash-Fin at the 2026 Inclusion·Conference on the Bund on September 10, 2026, positioning it as an open model grounded in real-world financial workflows and emphasizing utility and efficiency for complex tasks. The model was co-developed with leading financial instituti Built on a Mixture-of-Experts architecture, Ling-3.0-flash-Fin has 124 billion total parameters but activates only 5.1 billion per token, delivering the broad knowledge capacity of a large model with the low inference costs and high deployment efficiency of a compact model. Ant Group reports competitive results on benc

Ant Group announced the open-sourcing of Ling-3.0-flash-Fin at the 2026 Inclusion·Conference on the Bund on September 9, 2026, framing it as an open model grounded in real financial business needs. The press release emphasizes utility and efficiency for complex tasks in domains requiring trustworthy sources, consistent Built on a Mixture-of-Experts architecture with 124B total parameters and 5.1B active parameters per token, Ling-3.0-flash-Fin aims to deliver large-model knowledge capacity while keeping inference costs and deployment requirements compact. It targets information retrieval from authoritative sources, research reasoning

On August 28, Ant Group, together with CICC (China International Capital Corporation) and industry experts, released Ling-3.0-flash-Fin as the first finance-enhanced model from Ant's Bailing (Inclusion AI) team. It retains the Ling-3.0-flash architecture and long-context capability, preserving the 124B total / 5.1B act The article characterizes the approach as mirroring a "swap post-training, not weights" philosophy—keeping the base model and pushing it toward professional scenarios through financial continual pre-training and domain post-training. In an official demo, the model processed a Google 2026 Q2 financial spreadsheet contai

Ling 3.0 Flash Fin is a newly listed finance-domain hosted model from InclusionAI, available through OpenRouter as of August 27, 2026. The defining combination is long-context handling (262,144 tokens) paired with an explicit finance orientation, making it relevant for tasks that require reasoning across lengthy, jargo The model is positioned for financial text generation and analysis, including reviewing filings, summarizing quarterly performance narratives, comparing disclosures across periods, extracting risk factors, drafting analyst-style commentary, and answering questions over long reports, along with supporting domain-specifi

BenchLM's tracker page records Ling 3.0 Flash Fin as released August 28, 2026, listing it as a proprietary reasoning model with a 262K context window available via Vercel AI Gateway. The model is tracked but not yet publicly ranked, with only 3 sourced benchmark rows and most evaluation categories (coding, reasoning, k The page documents the API model ID as `inclusionai/ling-3.0-flash-fin` and confirms Vercel AI Gateway availability, but leaves fields such as knowledge cutoff, input modalities, output pricing, and maximum output length unsourced. Coverage is explicitly thin, with 0/3 verified agentic benchmarks, so the entry currentl

TechNode reported on August 28, 2026, that Ant Group has launched the finance-tuned Ling-3.0-flash-Fin model, retaining the Ling-3.0-flash architecture and its 124 billion total parameters and 5.1 billion active parameters. The model's weight release was scheduled for the following week, and the article is sourced via Ant Group is described as offering a one-month free API period through OpenRouter for finance professionals and developers, framing Ling-3.0-flash-Fin as a domain-specialized variant of the existing Ling-3.0-flash backbone. TechNode's launch-day coverage is light on deep architectural detail beyond confirming the param

Vercel announced that Ling 3.0 Flash Fin from Inclusion AI is now available on its AI Gateway and free to use through September 25, 2026. The model is a finance-focused variant of Ling 3.0 Flash, featuring a 256K-token context window, up to 32K output tokens, reasoning support, and function calling, designed for financ Developers must choose between two model IDs based on desired post-promo behavior: `inclusionai/ling-3.0-flash-fin` continues billing at the standard rate after September 25, while `inclusionai/ling-3.0-flash-fin-free` returns an error after the free period to prevent future charges. The model can also be configured fo

CreateWith's coverage explains that Vercel added Inclusion AI's finance-focused Ling 3.0 Flash Fin to its AI Gateway as a temporary free offering through September 25, 2026. The piece highlights the technical combination of a 256K-token context window, reasoning, function calling, and 32K-token maximum output as enabli The article frames the release as a dedicated domain-specific variant rather than a general-purpose model with a finance label applied at the gateway level, and suggests prototype use cases including document-review assistants, internal research tools, and financial data agents that compose reports from multiple tool c

The Puter Developer page documents Ling 3.0 Flash Fin as InclusionAI's (Ant Group) finance-domain fine-tune of Ling 3.0 Flash, developed with financial institutions and domain experts. It preserves the base Mixture-of-Experts architecture with 124B total parameters, about 5.1B active per token, and a 262,144-token context window. The model supports function calling for agent-style workflows calling external data or calculation tools, with a max output of 33K tokens and output speed of 326 tokens per second at 1.25s time-to-first-token. Puter lists the release date as August 27, 2026, with input pricing at $0.04 per million tokens and output at $0.12 per million tokens, and confirms finance-specific evaluation on FinFIRST, FinCRAFT, and SpreadsheetBench, where InclusionAI reported competitive performance with similarly sized and larger general-purpose models on tool-intensive financial tasks. InclusionAI notes that outputs require professional review and are not investment advice, targeting developers building investment research assistants and financial document analysis agents.

OpenRouter's listing confirms Ling 3.0 Flash Fin as a finance-focused mixture-of-experts model from InclusionAI, built on Ling 3.0 Flash with 5.1B active parameters out of 124B total, released on August 27, 2026 with a 262K context window. It is designed for real-world investment workflows requiring complex multi-step tasks and long-horizon planning and execution, while retaining general capabilities in reasoning, coding, and mathematics. The model is hosted across providers including NovitaAI and DeepInfra with distinct latency and throughput profiles. Auto benchmark results from OpenRouter show NovitaAI achieving 63.2% on GPQA Diamond and 65.3% on TAU-Bench, compared to 50.5% and 70.7% respectively under auto-routing, with DeepInfra scoring 15.9% on GPQA Diamond. DeepInfra delivered the best throughput at 191 tok/s with 0.77s P50 latency, while NovitaAI averaged 113 tok/s with 1.18s latency. Weighted average prices were $0.01473 per million input tokens and $0.1427 per million output tokens, reflecting cache hits and discounts across providers.

Ling-3.0-flash-Fin is the first finance-enhanced model in the Ant Ling family, developed by Ant Group with financial institutions and domain experts, extending Ling-3.0-flash through continued training on high-quality financial data. The MoE model has 124B total parameters with 5.1B activated and a 256K context window, targeting long-horizon financial agent workflows including retrieval, evidence review, calculation, modeling, and report preparation. Thinking mode is enabled by default, with recommended sampling settings of temperature=1.0, top p=0.95, and top k=20. The Hugging Face model card details financial-specific capabilities including source-grounded search (with the open-sourced FinFIRST benchmark), multi-document reasoning across annual reports and regulatory filings, valuation and spreadsheet workflows, and research-ready outputs organized for professional review. Evaluation covered FinFIRST, FinSearchComp Verified, FinCRAFT, Finance Agent, APEX-Agents, SpreadsheetBench, and τ³-Banking, where it proved competitive with similarly sized and larger general-purpose models, particularly on source selection and tool-intensive financial tasks. The BF16 checkpoint shares Ling-3.0-flash's architecture and is compatible with SGLang and vLLM runtimes.