Long-document understanding often requires reasoning over many visually rich pages, making inference costly and prone to context rot. In this work, we propose InSight-doc, an agentic visual perception framework that treats visual resolution as an adaptive reasoning-time resource. InSight-doc starts from low resolution and selectively zooms into high-resolution regions for finer evidence, without relying on any external retriever. To train such an agent, we construct an active-perception corpus of 17.9K high-quality SFT examples with region-level zoom-in trajectories, accompanied by 19.2K hard RL examples. Through SFT+RL, InSight-doc-8B improves the baseline by 4.3--16.4 accuracy points over document VQA benchmarks. On long documents, it reduces hallucination by more than 40% and inference latency by 41%--68% while maintaining an accuracy lead. Our code, datasets, and model are released at https://github.com/m-Just/InSight-doc .
Alex: Welcome to another episode of ResearchPod. Today, we're looking at a paper called "InSight-doc," which tackles a surprisingly tricky problem: how do you get an AI to read a long, complex document without losing the plot?
Sam: Is this about AI struggling with really long reports? Like, it reads the first few pages fine, but by the end it's forgotten what it was even looking for?
Alex: That's exactly it. When you feed a long document to one of these AI models, the standard approach is to convert every page into a high-resolution image and hand all of it to the model at once. And that creates two problems. First, it's slow—processing hundreds of high-detail images takes real computing power. Second, and more subtly, the model starts to lose track. The researchers call this "context rot."
Sam: Context rot. So the model isn't crashing—it's just... getting confused by too much information?
Alex: Right. Think of it like trying to memorize a phone book by reading every entry at full attention. By page fifty, the details from page one are already fading. The model's attention gets spread so thin across so many pages that it stops being useful. It starts filling in gaps with plausible-sounding but wrong answers—what researchers call "hallucinations."
Sam: So the problem isn't just speed. It's that too much detail actually makes the answers worse.
Alex: Precisely. And that's the insight the paper builds on. The solution they propose is called "agentic visual perception." The word "agentic" just means the model takes actions—it doesn't passively receive everything at once. Instead, it behaves more like a person skimming a report.
Sam: How so?
Alex: Imagine you're handed a two-hundred-page financial report and asked one specific question. You wouldn't read every word at full concentration. You'd flip through quickly, spot the relevant section, and only then slow down to read carefully. InSight-doc teaches the model to do exactly that. It starts with a low-resolution pass over the whole document—just enough to get the lay of the land—and then it zooms into specific regions only when it decides it needs the fine detail.
Sam: So it's treating visual resolution like a budget. Spend it only where it matters.
Alex: That's a good way to put it. The model scans the document, reasons about what it's looking for, and then makes what the paper calls a "tool call"—essentially a request to crop and enlarge a specific region of a specific page. It gets back a high-resolution image of just that patch, reads it, and continues reasoning.
Sam: But how does it know when to zoom in versus when it already has enough to answer?
Alex: That's where the training comes in. The researchers built a specialized dataset to teach the model a "zoom-in trajectory"—think of it as a worked example of good reading strategy. The trajectory shows the model the full path: here's the blurry overview, here's the reasoning step, here's the zoom, here's what you find, here's the answer.
Sam: So they're showing it how an expert would navigate the document, step by step.
Alex: Exactly. And they do this in two stages. The first stage is called supervised fine-tuning—the model watches expert examples and learns to imitate them. That's like a student shadowing a librarian, observing how they search efficiently. The second stage is reinforcement learning, where the model practices on its own and gets feedback based on whether its final answer is correct. That's the student going off and doing research independently, learning from their mistakes.
Sam: Show them the process, then let them practice until they're reliable.
Alex: Right. And the result of combining those two stages is that the model learns to weave thinking and action together. It doesn't just retrieve and report—it reasons, zooms, reads, updates its reasoning, and then answers. The paper calls this a "multimodal chain-of-thought," because the chain of reasoning includes both text and visual evidence.
Sam: Does that actually make the answers more accurate? Or is it mainly a speed improvement?
Alex: The paper suggests both. By focusing only on relevant patches rather than processing everything at once, the model avoids what they call "diluted attention"—where the signal gets lost in the noise. That focus leads to a meaningful drop in hallucination rates. And on the speed side, for long documents, they report cutting the time to generate an answer by up to sixty-eight percent compared to the high-resolution-everything approach.
Sam: That's a significant gap. Though I'm wondering—what's the catch? Can the model zoom into the wrong place and get stuck?
Alex: That's a fair concern, and the paper acknowledges it. Each zoom-in step adds a small amount of overhead—you're making multiple requests rather than one big batch. But the researchers argue that the savings from not processing every page at full resolution far outweigh those extra steps. The net result is still substantially faster and more accurate.
Sam: And I'd imagine if the model zooms into the wrong spot, it might just... not find what it needs and have to try again?
Alex: That's the idea. And this is actually where the design differs from an older approach. Many document AI systems use what's called an external retriever—a separate piece of software that tries to pick the relevant pages before the main model even sees them. The risk there is that the retriever makes a bad guess, and the model never gets to see the right information at all. InSight-doc removes that separation. The model itself decides where to look, which means its reasoning and its navigation are always in sync.
Sam: So it's not outsourcing the navigation to a separate system that might steer it wrong. The model is responsible for its own search.
Alex: Exactly. It stays grounded because it has to verify details through a visual crop before it commits to an answer. It can't just confabulate—it has to go look.
Sam: That's a meaningful shift. Instead of dumping everything on the model and hoping it copes, you're giving it a way to be selective and deliberate.
Alex: And that's the core contribution of the paper. It's not a faster processor or a bigger model—it's a smarter reading strategy. One that mirrors how a careful human reader actually works. Thanks for listening to ResearchPod.