InclusionAI
A finance-enhanced Ling-3.0-flash model from InclusionAI, released with BF16 weights under the MIT license.
Ling 3.0 Flash Fin has a credible open-weight finance specialization: its official Hugging Face card publishes BF16 weights under MIT, while the documented Flash base is a 124B-total/5.1B-active MoE. Vercel documents text-only input, a 262K shared context window, routed availability through Novita and DeepInfra, and reasoning configuration. This supports a moderate technical score, but not a claim of independently measured finance or general-task superiority. It is materially below Claude Opus 4.8 and GPT-5.5 on coding, reasoning, public adoption, and evidence depth: neither model-specific coding nor preference benchmark results are supplied. It is closer to the lower-evidence DeepSeek V4 Flash and Qwen preview anchors in public signal, but scores lower on demonstrated agentic work because those anchors have stronger catalog evidence. Text-only I/O places it below Gemini 3.5 Flash and multimodal leaders. MIT weights and a listed free OpenRouter variant make the cost proposition favorable, although official provider pricing is not established and hosted availability can vary by router. Developer experience is adequate through Hugging Face weights and API routing, but deployment guidance and the parameter figures are explicitly base-model evidence rather than Fin-specific proof. Treat finance suitability, latency, tool behavior, and safety controls as validation requirements before production use.
