A senior investment research agent developer removed long-maintained web scraping and source management scripts in favor of search APIs that support independent integration and return structured results (summaries + original paragraphs). This shift reflects a trend where developers decouple information retrieval to focus on core research logic and source verification, driven by the shutdown of Bing Search API and growing demand for real-time data in agents.
Structured Output
结构化输出
Forcing models to generate responses in predefined formats like JSON for easy programmatic parsing.
Curated Intelligence
DeepSeek has open-sourced its internal inference framework, DeepInfer, aiming to reduce computational costs by optimizing throughput and latency for large-scale model services. The framework integrates high-performance operators, speculative decoding, and SWA Bounded Replay, offering native compatibility with both NVIDIA and Huawei Ascend platforms. Additionally, the released deepseek-recipe component ensures that community deployments align perfectly with official API experiences.
Amazon Bedrock now offers native structured output support for all Anthropic Claude models, ensuring responses conform to defined JSON schemas. This update leverages constrained decoding to eliminate the need for custom parsing, retry logic, and post-processing, enhancing reliability in data extraction and automated workflows.
Cloudflare released 'cf', an open-source, agent-centric CLI designed with structured JSON output and command discovery to overcome the inconsistent naming and human-only formats of its legacy tool, Wrangler. Built in TypeScript and integrating Vite, cf aims to standardize cloud interactions for both developers and AI agents, with a planned 18-month transition period to deprecate Wrangler.
Artificial Analysis has launched the Decision Index v1.0, a new benchmark specifically designed for emerging 'decision models.' The index evaluates these models across five dimensions—classification, tool calling, knowledge, hallucination, and reasoning—to quantify their ability to output structured options with associated probabilities in a single call.
This article introduces 'System One' decision models like Jev, which generate structured outputs from fixed options in a single forward pass by masking the vocabulary, offering significant latency advantages over traditional multi-step LLM decoding. It provides a GitHub repository guiding developers through dataset construction, fine-tuning, and calibration to build low-latency, deterministic classification or decision-making models.
Microsoft introduces Microsoft-Decision-1, a new model purpose-built for fast decision-scoring that delivers top-tier latency and quality on structured tasks like routing and classification. It outperforms both general-purpose LLMs and other specialized decision models, now available in Microsoft Foundry with OpenRouter support coming soon.
OpenAI has released the Decisions API in public beta, designed to return typed answers such as probabilities, fixed choices, or rubric scores from text and images. It operates approximately 10 times faster than the Responses API and currently supports only the gpt-6-luna model. The tool aims to streamline content classification and request routing, requiring updated SDKs for integration.
LangChain has introduced Jev, a new agent architecture component designed to mitigate the latency and cost issues inherent in traditional LLM-based agent loops. By incorporating a 'System One' fast-intuition reasoning mechanism, Jev leverages structured outputs and tool calling to handle partial decision evaluations without requiring full large model intervention, significantly boosting agent execution efficiency. This marks a pivotal shift in agent development from sole reliance on LLMs toward hybrid cognitive architectures.
TypeSafe AI has released Jev, a novel model that outputs structured decisions for code execution instead of generating text. This approach replaces expensive general-purpose LLM calls by letting code own the workflow while AI handles narrow tasks, integrated with LangGraph for efficient production-grade agent deployment.