Analyzing Page Diffs at Scale with GPT-OSS 120B: The Page Deltas Story Powered by NLP Cloud

About Website Change Monitoring, Page Deltas, and NLP Cloud

The web changes constantly. Competitors adjust their pricing, vendors quietly update their terms of service, regulators publish new guidelines, and documentation pages evolve without any announcement. For many teams, discovering these changes days or weeks too late is expensive: a missed price move, an unnoticed policy update, or a silently deprecated API can have real business consequences. Page Deltas was built to solve exactly this problem. It continuously monitors the web pages that matter to a team, detects changes, and delivers clear, AI-generated summaries directly to Slack, Microsoft Teams, Discord, email, or webhooks.

Behind the scenes, the intelligence layer of Page Deltas is powered by GPT-OSS 120B, an open-weight large language model served on the NLP Cloud API. This partnership is a great illustration of what modern open-source AI makes possible: a production feature that reads thousands of page diffs every day, decides which ones actually matter, and explains them in plain language, all without sacrificing data privacy or cost control.

Why Analyzing Page Diffs at Scale Is Hard

Detecting that a page changed is the easy part. A simple comparison between two versions of a page flags something on almost every fetch: rotating banners, trending article widgets, cookie consent notices, updated timestamps, randomized markup generated by modern front-end frameworks, or A/B test variations. The hard part is deciding whether a change actually matters.

Traditional rule-based approaches like CSS selector ignore lists and change-size thresholds quickly reach their limits. They are brittle, they have to be maintained separately for every monitored website, and they fundamentally cannot tell whether a rewritten paragraph is a significant update to a refund policy or a harmless typo fix. Understanding a diff requires actually reading it, which is exactly what large language models are good at.

But at the scale of thousands of monitored URLs checked continuously, sending every candidate diff to an LLM is only viable if the model is accurate, fast, and cost-effective at the same time. This is where GPT-OSS 120B on NLP Cloud comes in.

Page Deltas Turns Raw Page Changes into Actionable Alerts

Page Deltas fetches the monitored pages on a schedule, compares each new version with the previous one, and captures before and after screenshots so users can visually verify any change. Its sitemap discovery feature even detects brand new URLs as soon as they are published, which is particularly useful for spotting product launches early. Teams can collaborate on monitors with role-based access, and route alerts to the channels they already use: Slack, Discord, Microsoft Teams, email, or webhooks for custom integrations.

Typical use cases include monitoring competitor pricing pages, tracking product releases and changelogs, watching regulatory and legal pages for compliance, and analyzing hiring pages to understand a market. In every case, the promise is the same: instead of a wall of raw HTML diffs, users receive a short, human-readable summary of what changed and why it might matter, only when it actually matters.

Filtering and Summarizing Page Diffs with GPT-OSS 120B

For every change detected by its monitoring engine, Page Deltas sends the diff, together with some page context, to GPT-OSS 120B through the NLP Cloud API. The model plays two roles at once. First, it acts as a filter: it evaluates whether the change is meaningful or just noise, so irrelevant updates never reach the user. Second, it acts as a writer: for meaningful changes, it produces the concise summary that lands in the alert, explaining in plain language what changed on the page.

GPT-OSS 120B is a particularly good fit for this workload. Released by OpenAI under the permissive Apache 2.0 license, it is a mixture-of-experts model: only around 5 billion of its 117 billion parameters are active for each token, which gives it near-frontier reasoning and instruction-following quality at the latency and cost of a much smaller model. Its large context window comfortably fits even big diffs together with the surrounding page content. For a high-volume, always-on task like page diff analysis, this balance of quality, speed, and price is exactly what is needed.

Serving GPT-OSS 120B on NLP Cloud brings additional benefits. The pay-as-you-go API scales with the volume of monitored pages without any infrastructure work on the Page Deltas side, and the data sent to the API is never used to train other models. And because GPT-OSS 120B is open-source, Page Deltas keeps full control over its roadmap: the model can later be fine-tuned on their own examples of relevant and irrelevant changes, or deployed on a dedicated server as volumes keep growing, all without vendor lock-in. Learn more about generative AI models like GPT-OSS 120B on NLP Cloud here.

What Is Next for Page Deltas and NLP Cloud

This collaboration is only getting started. As Page Deltas grows, the two teams are exploring deeper AI capabilities: richer categorization of changes, trend analysis across successive versions of a page, and monitoring of websites in many languages, an area where the multilingual abilities of the models served on NLP Cloud shine. On the NLP Cloud side, workloads like this one keep pushing us to optimize our inference stack so that large open-source models remain fast and affordable at scale.

Page Deltas is a great example of what developers can build today by combining a focused product with advanced open-source AI models served through a simple API.

If not done yet, you can sign up to NLP Cloud here.

You can try Page Deltas for free here.

Jessica
Head of Marketing at NLP Cloud