Iris: Climbing to the Search Frontier

• URL: http://arxiv.org/abs/2609.04304v1
• Date: Thu, 03 Sep 2026 17:51:01 GMT
• Title: Iris: Climbing to the Search Frontier
• Abstract summary: We present Iris-mini and Iris-pro, two search agents trained at the 35B-A3B and 397B-A17B scales, together with the data pipeline and training recipe behind them.
• Score: 21.066541702697112
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: We present Iris-mini and Iris-pro, two search agents trained at the 35B-A3B and 397B-A17B scales, together with the data pipeline and training recipe behind them. Tasks are reverse-constructed from the hyperlink structure of a web corpus: we author multi-hop chains over an entity graph distilled from a seed page and its out-links, rewrite every non-answer entity into a descriptive reference so that no clue can be resolved by string matching, and admit only questions that a reference model fails closed-book yet solves once the supporting evidence is supplied. These questions are then turned into trajectories, which are filtered at both the trajectory and the turn level before SFT. The policy is then optimized by RL against live search, with the reward judge and the observation summarizer served inside the training cluster, and with over-long rollouts interrupted at the request level and resumed from their committed prefix at the next step. We alternate the two stages in a procedure we call SFT-RL climbing, returning the hardest solved and most efficient rollouts of each RL round to the next supervised pass. Because inference-time context management is worth more on these benchmarks than most reported differences between systems, we evaluate every benchmark both with and without it, holding the tool set, the context limit, and the judge fixed. All results come from a single ReAct agent, with no sub-agents and no test-time verification. With management enabled, on BrowseComp, BrowseComp-ZH, DeepSearchQA, and HLE the two models reach $82.2/84.8/86.9/52.3$ and $88.6/85.1/92.9/56.4$, the strongest overall results among open-source search agents in their respective parameter ranges. We plan to release the model weights together with the complete recipe for data construction, training, and evaluation.

Related papers

• WeAgent-MMSearch: Native Text-Vision Interaction for Multimodal Search Agents [53.426679066400844]
  We introduce WeAgent-Harness, a multimodal agentic harness that supports native text-vision interaction and runtime recovery.<n>We develop WeAgent-MMSearch, an integrated system spanning data construction, agentic post-training, and multimodal rollout.<n>We also introduce VisTarget-Bench, a 150-task human-verified benchmark that pairs each question with a held-out target image.
  arXiv  Detail & Related papers  (2026-08-28T08:28:43Z)
• GTA-RAG: Graph-Trajectory-Augmented Reinforcement Learning for Multi-Turn Retrieval-Augmented Reasoning [27.83191719981676]
  We present textscGTA-RAG, a graph-trajectory-augmented RL framework for retrieval-augmented reasoning.<n>Method consistently outperforms RL-based RAG baselines with both Qwen2.5-3B and Qwen2.5-7B backbones.
  arXiv  Detail & Related papers  (2026-08-23T16:05:20Z)
• Search-on-Graph-R1: Training Large Language Models to Search Knowledge Graphs with Reinforcement Learning [44.1709905429432]
  Search-on-Graph-R1 internalizes navigation into a compact 8B model.<n>sogrone at 8B surpasses every frozen frontier-LLM system in our comparison.
  arXiv  Detail & Related papers  (2026-07-20T19:58:32Z)
• TRACE: Turn-level Reward Assignment via Credit Estimation for Long-Horizon Agents [51.523913302114266]
  TRACE (Turn-level Reward Assignment via Credit Estimation) is a dense credit-assignment method for agentic reinforcement learning.<n>It represents rollouts as state transitions at tool-call boundaries, obtains gold-answer log-probabilities from a frozen reference model, transforms them into log-ratio state values, and derives per-action rewards as Temporal-Difference changes in those values.<n>It substantially improves base-model tool-use ability using pure RL, without a cold-start supervised fine-tuning stage, an agentic mid-training stage, or training on live-web data.
  arXiv  Detail & Related papers  (2026-07-15T16:16:42Z)
• CoSearch: Joint Training of Reasoning and Document Ranking via Reinforcement Learning for Agentic Search [51.911048955965136]
  CoSearch is a framework that jointly trains a multi-step reasoning agent and a generative document ranking model.<n>Our results show that joint training of the reasoning agent and retrieval system is both feasible and strongly performant.
  arXiv  Detail & Related papers  (2026-04-19T17:48:17Z)
• PRAISE: Prefix-Based Rollout Reuse in Agentic Search Training [28.912613644535668]
  We present PRAISE, a framework for improving data efficiency and credit assignment in agentic search training.<n>Our method uses a single shared model for both search policy learning and prefix answer evaluation.<n>Experiments on multi-hop QA benchmarks show that PRAISE consistently improves performance over strong baselines.
  arXiv  Detail & Related papers  (2026-04-04T10:23:46Z)
• SWE-QA-Pro: A Representative Benchmark and Scalable Training Recipe for Repository-Level Code Understanding [41.98672557723593]
  We introduce SWEQA-Pro, a benchmark constructed from diverse, long-tail repositories with executable environments.<n>We also propose a scalable synthetic data pipeline that powers a two-stage training recipe: Supervised Fine-Tuning (SFT) followed by Reinforcement Learning from AI Feedback (RLAIF)<n> Empirically, a Qwen3-8B model trained with our recipe surpasses GPT-4o by 2.3 points on SWE-QA-Pro and substantially narrows the gap to state-of-the-art models.
  arXiv  Detail & Related papers  (2026-03-17T05:12:48Z)
• Compliance Brain Assistant: Conversational Agentic AI for Assisting Compliance Tasks in Enterprise Environments [2.8724171056550256]
  Compliance Brain Assistant (CBA) is a conversational, agentic AI assistant designed to boost the efficiency of daily compliance tasks for personnel in enterprise environments.<n>To strike a good balance between response quality and latency, we design a user query router that can intelligently choose between FastTrack mode and FullAgentic mode.
  arXiv  Detail & Related papers  (2025-07-23T07:51:10Z)
• Benchmarking Deep Search over Heterogeneous Enterprise Data [73.55304268238474]
  We present a new benchmark for evaluating a form of retrieval-augmented generation (RAG)<n>RAG requires source-aware, multi-hop reasoning over diverse, sparsed, but related sources.<n>We build it using a synthetic data pipeline that simulates business across product planning, development, and support stages.
  arXiv  Detail & Related papers  (2025-06-29T08:34:59Z)
• List-aware Reranking-Truncation Joint Model for Search and Retrieval-augmented Generation [80.12531449946655]
  We propose a Reranking-Truncation joint model (GenRT) that can perform the two tasks concurrently. GenRT integrates reranking and truncation via generative paradigm based on encoder-decoder architecture. Our method achieves SOTA performance on both reranking and truncation tasks for web search and retrieval-augmented LLMs.
  arXiv  Detail & Related papers  (2024-02-05T06:52:53Z)
• Boot and Switch: Alternating Distillation for Zero-Shot Dense Retrieval [50.47192086219752]
  $texttABEL$ is a simple but effective unsupervised method to enhance passage retrieval in zero-shot settings. By either fine-tuning $texttABEL$ on labelled data or integrating it with existing supervised dense retrievers, we achieve state-of-the-art results.
  arXiv  Detail & Related papers  (2023-11-27T06:22:57Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.