According to CoinMeta, Scaledown has announced the launch of a constrained decoding model ( CDMs ), which is a task-specific model focused on classification, extraction, and summarization, completely skipping the generation loop. This model reads input once and directly returns structured output, being 200 times faster than the leading LLM API and 467 times less costly, supporting multimodal contexts with up to 1 million tokens. It is currently available on platforms such as OpenRouter, Vercel, and Langchain, and is expected to change the company's internal deployment of the AI token economy model.