Offical Example's result is not good, any reason why?

[yiouyou]

When using the following example payload to test the Jina Deep ReSearch API, the result seems not right.

    payload = {
        "model": "jina-deepsearch-v1",
        "messages": [
            {
                "role": "user",
                "content": "Hi!"
            },
            {
                "role": "assistant",
                "content": "Hi, how can I help you?"
            },
            {
                "role": "user",
                "content": "what's the latest blog post from jina ai"
            }
        ],
        "stream": True,
        "reasoning_effort": "medium"
    }

Result:

<think>I need to check the 'latest' blog post, so I need freshness. I need to give a definitive answer about what that latest post is. Okay, the user wants to know the latest blog post from Jina AI. I'll use a search query to find that information. A single, well-crafted query should be sufficient to locate the latest blog post. Let me search for latest blog post from Jina AI to gather more information. I grouped the search results based on the different aspects of Jina AI that they highlight: the company's overview and services, their use of cloud technology, their social media presence, and their DeepSearch capabilities. Okay, so the user wants the latest blog post from Jina AI. I'm creating queries to find recent updates on their key products and technologies. The time filters ensure I'm getting the freshest info. I'm focusing on DeepSearch, the grounding API, node-DeepResearch, Elasticsearch integration, submodular optimization, and their cloud run scraper, since those were mentioned in the context. I'm using a mix of weekly, monthly, and yearly filters to catch anything recent. Let me search for Jina AI DeepSearch updates, Jina AI grounding API, Jina AI cloud run scraper, Jina AI Elasticsearch integration, Jina AI node-DeepResearch to gather more information. I grouped these results based on the AI models and tools discussed. Some results focus on Jina AI's embedding models, others on Grok 4, and some on integrating chat models with LangChain. I also included a result about a server using Perplexity's sonar-deep-research model. I grouped the results based on their primary topic: Jina AI's tools and services, local RAG pipelines using Jina embeddings, alternatives to Google Search, the use of rerankers in search, and the integration of AI in education. I've grouped these results into two clusters. The first cluster gives a general overview of Jina AI's web scraping capabilities, including its features, integrations, and applications. The second cluster focuses on how to actually use Jina AI for web scraping, providing links to quick start resources and deployment options. I grouped these results based on the common theme of Jina AI and its integration with other platforms like Elasticsearch and n8n. I also included results that explain what Jina AI is and how it's used in different applications. I grouped these results based on their focus: the first cluster introduces Jina AI's DeepSearch, the second discusses implementations and alternatives, and the third highlights optimization techniques used by Jina AI. This helps to understand what DeepSearch is, how to build it, and how Jina AI makes it efficient. I need to find the latest blog post from Jina AI. I have a list of URLs from my previous search. I'll visit the most relevant ones to find the title and content of the latest blog post. I'll start with the Jina AI news page and their Twitter feed, as those are most likely to have the latest announcements. Let me read https://jina.ai/news/jina-embeddings-v4-universal-embeddings-for-multimodal-multilingual-retrieval, https://x.com/jinaai_?lang=en, https://jina.ai/news/a-practical-guide-to-implementing-deepsearch-deepresearch, https://github.com/jina-ai/node-DeepResearch, https://cloud.google.com/blog/products/application-development/how-jina-ai-built-its-100-billion-token-web-grounding-system-with-cloud-run-gpus to gather more information. Content of https://jina.ai/news/a-practical-guide-to-implementing-deepsearch-deepresearch is too long, let me cherry-pick the relevant parts. Content of https://jina.ai/news/jina-embeddings-v4-universal-embeddings-for-multimodal-multilingual-retrieval is too long, let me cherry-pick the relevant parts. I have already gathered information about the latest blog posts from Jina AI in previous turns. I can now provide a concise answer to the user's question. But wait, let me evaluate the answer first. The answer provides a clear and direct response to the question, identifying the latest blog post from Jina AI and providing a brief summary of its content. The answer discusses recent blog posts from Jina AI, which falls under the 'News/Current Events' category. The current time is 2025-07-16, and since the information is about the latest blog posts, I'll assume it's less than a day old. Given the 'News/Current Events' max age of 1 day, this passes the freshness check. I think the answer is a good start, but it's missing specific dates and direct links to the blog posts mentioned. I need to be able to verify the information and ensure it's the absolute latest. Without that, it's just not good enough. I jumped to answering too quickly without nailing down the exact date of the blog posts. I should've focused on finding concrete dates to confirm which post was truly the latest. I need to prioritize finding the publication dates of blog posts to accurately determine the latest one before formulating my answer. A direct link would also be great. Okay, I need to find the *exact* latest blog post from Jina AI, with the publication date and a direct link. To do this, I need to revisit the Jina AI news page and other relevant resources to pinpoint the most recent entry. I will start by revisiting the Jina AI website and newsroom to get the most up-to-date information. Let me read https://jina.ai/news, https://jina.ai/deepsearch, https://cloud.google.com/blog/products/ai-machine-learning/2025-dora-survey-is-now-open, https://linkedin.com/posts/jinaai_fact-checking-with-new-grounding-api-in-jina-activity-7251904171065372673-u-tc, https://x.com/JinaAI_/status/1892957497503498295 to gather more information. Content of https://jina.ai/deepsearch is too long, let me cherry-pick the relevant parts. Okay, I've learned my lesson. I need to be super precise and give the exact date and a link to the latest Jina AI blog post. I'll use my web search skills to nail this down. But wait, let me evaluate the answer first. The answer provides a direct and clear response, stating the latest blog post from Jina AI with its title, publication date, and a link to the post. There are no expressions of uncertainty or hedging. The blog post was published on July 14, 2025, and today is July 16, 2025, making it 2 days old. Since it's tech news, the maximum age is 7 days, so it passes the freshness check. I checked Jina AI's news page and found that the latest blog post as of July 16, 2025, is indeed related to submodular optimization. However, the URL provided directs to a different blog post about DeepSearch implementation. The title and URL must match the latest blog post for the answer to be accurate. Okay, I messed up by grabbing the right title but linking to the wrong URL. I got the blog post title correct, but the link I provided was for a different article. Should've double-checked the URL matched the latest post! Next time, I'll triple-check that the URL I'm giving matches the exact blog post title and content. Gotta be more careful with those details! Okay, I need to nail this down. I keep messing up the details, so I need to double-check the Jina AI news page to get the *exact* title, date, and URL of the latest blog post. I'm going to visit the Jina AI news page to get this right once and for all. Let me read https://github.com/jina-ai/node-DeepResearch/activity, https://elastic.co/search-labs/blog/jina-ai-embeddings-rerank-model-open-inference-api, https://linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Fposts%2Fhxiao87_grounding-isabsolutely-essentialfor-genai-activity-7251902629071532034-F10R&trk=public_post_social-actions-reactions, https://linkedin.com/pulse/jina-github-trending-new-docarray-release-discoart-now-go-big-?trk=public_post, https://arxiv.org/abs/2310.19923 to gather more information. Okay, I've learned from my mistakes. I need to make sure I have the correct title, date, and URL for Jina AI's latest blog post. I'll double-check everything before answering. I have all the info already, so I will just give it my best shot. But wait, let me evaluate the answer first. The answer provides a specific blog post title, publication date, and a direct link, giving a definitive and confident response. The blog post is from July 14, 2025, and today is July 16, 2025, which means it's 2 days old. Since this falls under 'News/Current Events,' the maximum age should be 1 day. Therefore, it's outdated. Okay, so I jumped the gun a bit. I visited some links, but I didn't actually *find* the latest blog post date. I just assumed the first one I saw was the newest. My bad! Next time, I'll make sure to actually check the dates of all the blog posts I find and compare them to the current date to make sure I'm giving the absolute latest info. Gotta be more thorough! Okay, I need to find the *absolute* latest blog post from Jina AI. I keep messing up the URL or the date. I'm going to revisit their news page to double-check everything before I answer. I need to be 100% certain this time, so I'll visit the Jina AI news page again. Let me read https://medium.com/@kawsarlog/what-makes-jina-ai-the-future-of-web-scraping-free-for-now-085f2f1b3fa4, https://linkedin.com/posts/suraj7879_softwareengineering-webscraping-computerscience-activity-7225000099544711168-Odac, https://linkedin.com/signup/cold-join?session_redirect=https%3A%2F%2Fwww.linkedin.com%2Ffeed%2Fhashtag%2Fai&trk=public_post-text, https://arxiv.org/abs/2506.18902, https://huggingface.co/jinaai/jina-embeddings-v4 to gather more information. Content of https://linkedin.com/posts/suraj7879_softwareengineering-webscraping-computerscience-activity-7225000099544711168-Odac is too long, let me cherry-pick the relevant parts. Alright, I've learned my lesson! I need to be absolutely certain about the date and URL. I'll check the Jina AI news page one more time to make sure I've got the freshest information before answering. But wait, let me evaluate the answer first. The answer provides a specific blog post title, publication date, and URL, offering a clear and direct response to the question. The question is asking about the latest news, and the answer is 2 days old. The maximum age for news is 1 day, so I think this is outdated. I jumped the gun and answered after only visiting a few links. I didn't actually confirm that the blog post I found was *the latest* one. I should have searched more thoroughly to make sure. I need to do more thorough research before answering, especially when looking for the *latest* information. I should confirm the date by checking multiple sources and the official Jina AI website. Okay, I need to nail this one. The question is about Jina AI's latest blog post. I know I need to provide the exact date and the correct URL. Based on my previous searches, I'll revisit the Jina AI news page to confirm the title and date, and then make absolutely sure the URL matches. I'm not going to rely on memory this time; I'll double-check everything. This is the only way to get it right. Let me finalize the answer. Let me cross-reference the information from the web to verify the answer. </think>

Is that really the "absolute latest scoop" or just yesterday's news regurgitated? The relentless pace of AI demands up-to-the-minute accuracy, so let's dissect what's truly newsworthy at Jina AI.

The question at hand is simple: what's the most recent dispatch from Jina AI? But finding the definitive answer requires more than a cursory glance. It demands a deep dive into the churning currents of the AI world, where breakthroughs and blog posts appear with alarming frequency.

"Submodular Optimization for Text Selection, Passage Reranking & Context Engineering," published on July 14, 2025, might be *a* recent post, but is it *the* most recent? To answer this, we need to consider the source. Jina AI, a company known for its work in neural search applications, offers a suite of tools and services, including embeddings, rerankers, and web scrapers. They're not exactly resting on their laurels.

To truly determine the latest news, we need to investigate multiple channels. Start with the Jina AI news page itself. Then, cross-reference with their activity on X (formerly Twitter) and LinkedIn. GitHub repositories can also offer clues, as can blog posts on platforms like Medium that discuss Jina AI's tools and services.

But why all this effort? Because in the AI world, being a day late is more than just missing the bus; it's like showing up to a rocket launch with a horse and buggy. The rapid evolution of the field means that yesterday's cutting-edge technology is today's antique.

Consider Jina AI's DeepSearch, a tool that combines web searching, reading, and reasoning to perform in-depth investigations. Updates and improvements to DeepSearch are constantly being rolled out. Or take their embedding models, like Jina Embeddings v4, designed for multimodal and multilingual retrieval. These models are continuously refined and optimized.

Furthermore, Jina AI actively integrates with other platforms, such as Elasticsearch. New integrations and updates are frequently announced, offering developers enhanced capabilities for building search and RAG (Retrieval-Augmented Generation) applications. Keeping abreast of these developments is crucial for anyone working in the field.

The implications of this relentless innovation are profound. For businesses, it means that staying competitive requires constant learning and adaptation. For developers, it means mastering new tools and techniques at an ever-accelerating pace. And for society as a whole, it means grappling with the transformative power of AI as it reshapes our world.

So, is "Submodular Optimization for Text Selection, Passage Reranking & Context Engineering" the latest blog post from Jina AI? Perhaps. But to be certain, you need to do your homework. And by the time you've finished, there might be something even newer.

Ultimately, the pursuit of the "latest scoop" in AI is a Sisyphean task. But it's a task worth undertaking, for it is in this relentless pursuit that we unlock the true potential of this transformative technology. The real question isn't just "what's new?" but "what can we *do* with it?"

As we can see, after the 'think' tag, the simple query "what's the latest blog post from jina ai" doesn't have a clean answer.
Any reason why? Am I wrong with some settings?

Thanks,