ERNIE 5.1
ERNIE 5.1 is Baidu's latest ERNIE model, derived from ERNIE 5.0 with total parameters cut to about one-third and active parameters to about one-half; Baidu says it took about 6% of the pre-training cost of comparable models. It is served through Baidu's Qianfan platform with a 128k context.
Released
May 9, 2026
Type
llm
License
proprietary
Context
—
Baidu lists a “128k” context window, with up to 119k input and 65,536 output tokens.
Pricing
Priced in Chinese yuan per thousand tokens on Baidu’s Qianfan platform, so no per-million figure is shown: ¥0.004 input / ¥0.018 output per 1K tokens for input up to 32K, and ¥0.006 / ¥0.022 for input from 32K to 128K. No US dollar price list was found.
As the developer's own pages state it, checked . Not listed on the pricing page, which compares US dollar prices per million tokens.
Capabilities
Sources
Every figure on this page was checked against these pages on . A figure we could not source is left out.
- ERNIE Blog: ERNIE 5.1 Officially Released (May 9, 2026) (opens in a new tab)
- Baidu AI Cloud Qianfan: Model list (Chinese, updated 2026-10-14) (opens in a new tab)
- Baidu AI Cloud Qianfan: Model service billing (Chinese, updated 2026-10-09) (opens in a new tab)
Spotted something out of date? Tell us.
Links
In the news: ERNIE 5.1 and Baidu
Search Engine Journal
Baidu Now Measures AI Citations: Here’s What SEOs Should (And Shouldn’t) Read Into It via @sejournal, @MPentzek
Baidu has introduced a new metric within its AI Traffic Analysis report that quantifies artificial intelligence-driven clicks based on citations rather than traditional impressions. This significant shift in measurement methodology means that the report will now distinguish between users who arrive at a website after seeing a link within an AI-generated response and those who might have encountered the link through other means or simply viewed it without clicking. The AI Traffic Analysis report, as detailed by Search Engine Journal, aims to provide search engine optimization (SEO) professionals with a more granular understanding of how AI is influencing website traffic and user engagement. Historically, web analytics have relied heavily on impression data, which counts how many times a link or advertisement is displayed to a user, and click-through rates (CTR), which measure the percentage of impressions that result in a click. Baidu's new approach moves beyond this by specifically isolating and measuring traffic that originates from direct citations within AI-generated content. This implies that the AI is not just presenting information but is actively directing users to specific sources, and Baidu is now tracking the effectiveness of these AI-driven referrals. The implications for SEO are substantial, as it suggests a need for strategies that optimize content for direct citation within AI outputs, potentially influencing how websites are ranked and presented by AI models. Search Engine Journal highlights that this new metric is crucial for understanding the evolving landscape of search and content discovery. As AI models become more sophisticated and integrated into search experiences, the way users find information is changing. Baidu's report offers a tangible way to measure the impact of these AI-driven interactions. SEO professionals will need to analyze this data to understand which types of content are most likely to be cited by AI, and how to improve their visibility and authority in AI-generated search results. This could involve focusing on creating authoritative, well-structured content that AI models can easily understand and reference, as well as building strong backlink profiles that signal credibility to AI systems. While the report provides valuable new data, Search Engine Journal also cautions SEOs against overinterpreting certain aspects. The focus on citations is a specific lens through which to view AI traffic, and it does not negate the importance of other SEO factors. Traditional SEO practices, such as keyword optimization, technical SEO, and user experience, will likely remain critical. The AI Traffic Analysis report should be viewed as a complementary tool, offering insights into a particular facet of AI's influence on search rather than a complete overhaul of SEO strategy. The report's ability to differentiate between AI-driven citations and other forms of traffic is its key innovation, allowing for more precise analysis of AI's role in directing users to web content.
The Hacker News
CDN Tsunami Attack Abuses HTTP/3 Translation for Up to 350x DoS Amplification
Cybersecurity researchers have detailed two denial-of-service (DoS) attacks, collectively named "CDN Tsunami," which leverage the translation process between HTTP/3 and HTTP/1.1 protocols used by major content delivery networks (CDNs). These attacks can amplify a low-bandwidth request into a significantly larger one directed at the origin server, achieving amplification factors of up to 350x. The researchers successfully tested these exploits against CDNs operated by Alibaba and Baidu, demonstrating their real-world applicability and potential impact on web infrastructure. The core of the attack lies in how CDNs, designed to improve website performance and scalability, handle traffic. When a client connects using the newer HTTP/3 protocol, the CDN's edge servers often translate these requests into the older HTTP/1.1 protocol before forwarding them to the origin web server. This translation process, particularly the way certain headers and request components are handled, can be manipulated to create a disproportionately large response from the origin server for a small initial request. The "CDN Tsunami" attacks exploit this by crafting specific HTTP/3 requests that, upon translation to HTTP/1.1, trigger excessive resource consumption on the origin server. This can manifest as a flood of data or computational load, overwhelming the server and leading to a denial of service for legitimate users. The researchers highlighted that the effectiveness of these attacks is dependent on the specific implementation of HTTP/3 to HTTP/1.1 translation by the CDN provider. Different CDNs may have varying degrees of vulnerability to this type of manipulation. The findings underscore a critical security consideration for CDNs and the organizations that rely on them for traffic management and performance optimization. As web traffic increasingly adopts newer protocols like HTTP/3, it is imperative for CDN providers to rigorously audit their protocol translation mechanisms for potential security weaknesses. The research team has not yet publicly disclosed the full technical details of the exploits, likely to allow CDN providers time to implement necessary mitigations. However, the disclosure serves as a significant warning to the cybersecurity community about a novel attack vector targeting a fundamental component of modern web delivery. The implications extend to any service that relies on CDNs for its online presence, as a successful DoS attack can lead to significant downtime, financial losses, and reputational damage. The researchers' work emphasizes the ongoing need for vigilance and proactive security measures in the face of evolving network protocols and attack methodologies. The ability to achieve such high amplification factors with relatively low-bandwidth attack traffic makes these exploits particularly potent, as they can be launched from less powerful infrastructure.
Electrek
Baidu’s Apollo Go begins London robotaxi testing with Uber and Lyft
Baidu's autonomous driving unit, Apollo Go, initiated robotaxi road testing in London on March 18, 2024, marking its entry into the European market. This expansion into the UK capital sees Apollo Go collaborating with two prominent ride-hailing platforms simultaneously: Uber and Freenow, which is operated by Lyft. This dual partnership signifies a significant step for Baidu's global ambitions in autonomous mobility. The commencement of testing in London follows closely on the heels of another major milestone for Apollo Go, which successfully conducted the first fully driverless trial in a right-hand-drive market in Hong Kong just days prior. This rapid succession of international deployments highlights Baidu's accelerated pace in testing and validating its autonomous driving technology across diverse regulatory and operational environments. Baidu's Apollo program is a comprehensive open-source platform for autonomous driving, aiming to foster collaboration and accelerate the development of self-driving technology. Apollo Go represents the commercial application of this platform, focusing on ride-hailing services. The technology underpinning Apollo Go includes advanced sensors such as lidar and radar, sophisticated AI algorithms for perception and decision-making, and high-definition mapping. The service has previously operated in various Chinese cities, including Beijing, Shanghai, and Guangzhou, as well as in Japan. The introduction of Apollo Go in London is subject to regulatory approvals and will initially involve safety drivers, with plans to transition to fully driverless operations as regulations permit. The company has stated its commitment to adhering to stringent safety standards and collaborating with local authorities throughout the testing phase. The London testing grounds will allow Apollo Go to gather crucial data on performance in a complex urban environment characterized by varied weather conditions, dense traffic, and intricate road layouts. This data will be instrumental in refining the AI models and operational strategies for autonomous vehicles. The partnership with Uber and Freenow provides Apollo Go with access to extensive user bases and established operational networks, facilitating a smoother integration into the local transportation ecosystem. For Uber and Freenow, this collaboration offers an opportunity to explore the integration of autonomous vehicles into their existing ride-sharing services, potentially leading to more efficient and cost-effective mobility solutions in the future. The competitive landscape for robotaxi services is intensifying globally, with companies like Waymo and Cruise also expanding their operations in various cities. Baidu's entry into London positions it as a key player in this burgeoning industry, aiming to replicate its success in China on a global scale.
MarkTechPost
How to Build an End-to-End OCR Pipeline with Baidu’s Unlimited-OCR for High-Resolution Images and Multi-Page PDF Parsing
A tutorial published this week outlines the construction of a comprehensive Optical Character Recognition (OCR) pipeline utilizing Baidu's Unlimited-OCR model. This workflow is designed to process both high-resolution document images and multi-page PDF files, enabling structured data extraction. The process begins with configuring a GPU environment and installing necessary dependencies, including specific versions of libraries like `transformers==4.57.1` and `pymupdf`. The tutorial details loading Baidu's 3-billion parameter vision-language model, with automatic selection between bfloat16 or float16 data types based on GPU support. It confirms GPU availability and identifies the specific GPU model in use, such as `NVIDIA GeForce RTX 3090`. The model is then loaded and moved to the GPU for inference. The workflow evaluates two inference modes for single-page OCR: the tiled Gundam mode and the faster Base mode. Following single-page processing, the pipeline is extended to handle multi-page PDF parsing. This involves using the `PyMuPDF` library and the `infer_multi()` function. Throughout the entire process, the tutorial emphasizes maintaining long-context generation settings, repetition controls, and structured output handling. These features are crucial for accurately processing dense document layouts, tables, paragraphs, and content that spans across multiple pages, ensuring a reproducible end-to-end extraction pipeline. The tutorial provides code snippets for dependency installation and model loading, demonstrating the practical steps involved. It highlights the model's capability to manage complex document structures and cross-page information, making it a robust solution for document analysis tasks. The use of `torch.cuda.is_available()` and `torch.cuda.get_device_name(0)` ensures compatibility and optimal performance on available hardware.
Electrek
Baidu Apollo Go wins Level 4 robotaxi approval in Switzerland as AmiGo
Baidu's Apollo Go secured a Level 4 autonomous driving permit in Switzerland for its AmiGo robotaxi service on March 11, 2024, marking a significant step in European public transport deployment. This permit, granted by the Federal Roads Office (FEDRO), allows operations within an 80 km² service area spanning three eastern cantons. Baidu, in partnership with Swiss Post's PostBus, aims for AmiGo to become Europe's largest planned automated public transport operation. The company stated that this approval positions them ahead of competitors like Waymo and Tesla in bringing driverless vehicles to public transit networks on the continent. The AmiGo service is expected to commence operations in the coming months, following extensive testing and regulatory review. This development signifies a growing international acceptance and integration of autonomous vehicle technology into urban mobility solutions.