agentic-search

Mixedbread introduces Toast 1, a specialized search agent that matches frontier model quality while being up to 10x cheaper and 12x faster. It automates agentic search loops and achieves state-of-the-art results on benchmarks like OfficeQA Pro V2 and legal knowledge tasks.

Samsja19 highlights their collaboration with Mixedbread AI on Toast 1, a specialized agentic search model that achieves frontier quality at 12x faster speed and 1/10th the cost via reinforcement learning.

AgonAlpha is a new autonomous alpha discovery system that searches over verified research artifacts and uses an adversarial reviewer with veto authority plus pending-aware budget allocation. Deployments on WorldQuant BRAIN achieved high fitness and Sharpe scores while preserving full provenance.

Neon and Castform team up to show how RL post-trained open-weights models combined with Lakebase Search can beat frontier models like GPT-5.6 Sol on retrieval tasks at 100x lower cost and latency.

SimpleWikiSearch 是一个可复现的离线 Wikipedia 环境，用于评估基于 LLM 的 agentic 搜索系统，明确规定了语料构建、检索栈和工具接口。

A controlled scaling study of retrieval-augmented generation paradigms finds that BM25 lexical retrieval outperforms agentic and graph-based retrieval at scale, while agentic search only leads on small corpora.

XYZ AI Lab releases XYZ-Aquila-mini, an open-weight thinking model for agentic deep search, fine-tuned from Qwen3.6-35B-A3B, achieving top benchmark scores among sub-40B open-weight models.

This paper introduces the Relevance-Aware RipGrep Search Agent (RARG), which uses relevance as an execution prior for corpus interaction in agentic search, improving accuracy and efficiency over existing methods.

This paper introduces Multi-Agent Protocol Distillation (MAPD), a framework that distills knowledge from proprietary agentic search models into open-source models using a structured protocol to bridge the distribution gap, achieving state-of-the-art results on seven QA benchmarks.

XYZ AI Lab releases XYZ-Aquila-pro, an open-weight thinking model for agentic deep search, post-trained from Qwen3.5-397B-A17B via a bounded-exploration AI4AI pipeline, with strong benchmark results in sub-400B open-weight comparisons.

A benchmark test of nine agentic web search tools using Claude Opus 4.8 agent found that lower-cost tools often outperformed more expensive ones, with Firecrawl ranking first for accuracy and Serper being the most cost-effective.

Auto-FL-Research introduces a constrained coding-agent workflow for automatically searching and evaluating federated learning algorithmic recipes, showing performance gains on multiple healthcare and LEAF tasks while also exposing seed-sensitive and search-selected failure cases.

LlamaIndex introduces agentic retrieval with LlamaParse Index, combining semantic search and grep for flexible agent harnesses. A webinar on June 30th will demonstrate these tools.

This paper identifies an anchor collapse phenomenon in agentic search where parallel trajectories converge due to similar initial queries, and proposes DivInit, a training-free method that samples diverse initial queries to improve multi-hop question answering performance.

This paper introduces DR-DCI, a retriever-steered framework for scaling direct corpus interaction by dynamically expanding a local workspace, achieving improved accuracy and efficiency in agentic search over large corpora.

Describes improving agentic memory search by incorporating grep-based exact matching alongside vector embeddings, inspired by a paper; achieved significant recall gains in their memory layer.

The author describes using a knowledge graph extractor built with a Qwen model to generate challenging multi-hop QA pairs for evaluating agentic search systems.

This paper from eBay presents a modular two-agent simulation framework for evaluating conversational shopping assistant architectures, enabling controlled comparisons of responder designs. Key findings include that rolling-window memory outperforms intent-extraction memory by 35% in speed, and that systematic failure analysis reduced failure rates by 62%.

Fable-5/Mythos achieves new SOTA on agentic search but is expensive for self-hosting, while open-weight Harness-1 offers a cost-effective alternative with fewer query restrictions.

This empirical study compares grep and vector retrieval strategies in LLM agent workflows, finding that grep generally yields higher accuracy across different agent harnesses and tool-calling styles, with performance heavily dependent on harness choice and context engineering.