LangChain adds ScaleDown constrained decision models for agent workflows
Article timeline
The articles in this story. Times are ET.
Rohan Paul@rohanpaul_aiXScaleDown releases three small constrained decoding models for classify, extract, and summarizeAIScaleDown releases Classify, Extract, and Summarize, three Constrained Decoding Models that select answers from the input rather than generating them token by token. The models accept up to 1M tokens of text and images, score 77.2% on DecisionBench versus 72% for Jev, and run 200x faster and 467x cheaper than frontier LLM APIs. Pricing is $0.05 per million input tokens with output free.
elvis@omarsar0XScaleDown launches constrained decoding models for classify, extract, and summarize tasksAIScaleDown releases Constrained Decoding Models, which classify, extract, or summarize 1M tokens of text and image input without generating output token by token. The company says the models run 200x faster and 467x cheaper than frontier LLM APIs, at $0.05 per million input tokens with free output. It reports 77.2% on DecisionBench against 72% for Jev and is available through OpenRouter, Vercel, LiteLLM, LangChain, and other platforms.
LangChain@LangChainOfficialLangChain partners with ScaleDown to integrate constrained decision models into agentsAILangChain says ScaleDown's constrained decision models can now plug into agents through langchain-scaledown, an open-source package with prebuilt middlewares. The models are also supported in the LangSmith LLM Gateway, which adds tracing, spend limits, and rate limits.
- LangChain@LangChainOfficialLangChain adds support for ScaleDown's constrained decision models
AILangChain now supports constrained decision models from ScaleDown, which the company says are optimized for quick, cost-effective decisions. The models are compatible with LangChain and Deep Agents, and calls can be traced with LangSmith LLM Gateway. They are aimed at summarizing, extracting, and classifying.
- LangChain BlogOfficialLangChain adds ScaleDown models for managing long-running agent context
AILangChain has released a langchain-scaledown integration and LangSmith LLM Gateway support for ScaleDown's constrained decoding models. The integration ships three middlewares for LangChain and Deep Agents that summarize old history, compress large tool results, and extract structured data. ScaleDown reports its models are 200x faster and 467x cheaper than frontier models on these narrow tasks.