Baidu introduces updated search API with self-learning and lowers cost to 4.9 yuan per 1000 queries
Read more
Pandaily
pandaily.com

Baidu introduces updated search API with self-learning and lowers cost to 4.9 yuan per 1000 queries

At a product briefing for the Baidu application in Beijing on October 30, the company announced that its search engine has achieved what is called recursive self-evolution. Additionally, an updated search AI API was launched at a promotional price, which Baidu claims is lower than DeepSeek's search API.

Self-evolution is described as a closed loop involving information extraction, feedback, and retraining. Every real search and subsequent user feedback are used as training signals and fed back into the system, allowing the models to continuously improve without manual intervention. Baidu asserts that thanks to this online feedback loop, several agents improve collaboratively, and the overall efficiency of iterations has increased by more than 80 percent. The company has not published independent evaluations of this claim.

The API is based on a four-layer architecture, which Baidu calls AI-native. The content layer uses data from over 80 languages and an index containing hundreds of billions of high-quality pages, with data reliability increasing by more than 50 percent, according to the company. The self-evolution mechanism functions at the engine level. The architectural layer ensures low latency and high concurrent request processing for enterprise use. The application supports access via API, MCP, and a set of skills. Baidu reports that the quality of its own assistant has improved by 20 percent after integration.

According to the Baidu AI Cloud campaign page, the enhanced search package costs 4.9 yuan per 1000 queries under a promotion running from October 30 to December 31, compared to the regular price of 40 yuan. Buyers can make repeat purchases until reaching a cumulative limit of 1 million queries, after which the standard price applies. The flagship package will cost 9.9 yuan per 1000 queries, which is lower than the initial price of 72 yuan, and requires an invitation code. Results provided include the website, title, URL, and source snippet for verification purposes.

A consumer version of the application was also presented during the briefing. Baidu notes that search is transforming from a 'one question—one answer' format to a 'one task—one result' format. Complex queries are processed through multi-step searches that provide a verified summary; routine tasks are executed directly, and long-term topics can be tracked using user-consented notifications. This functionality covers over 170 use cases. Interactive result pages now appear in five times more scenarios and utilize over 5 billion data resources from government agencies, utilities, and the central bank. Furthermore, the AI camera feature can now simultaneously track multiple objects and hold a live dialogue in nine Chinese dialects, including Northeastern, Sichuanese, and Cantonese.

Similar stories

Google's Financial Results: Cloud Segment Growth Contrasts with Search Revenue Slowdown Due to AI Investments
Read more
techcentral.co.za

Google's Financial Results: Cloud Segment Growth Contrasts with Search Revenue Slowdown Due to AI Investments

Alphabet, Google's parent company, reported disappointing financial results in its core search advertising business, alongside a sharp increase in cash burn rates amid massive investments in artificial intelligence development.

Despite this, Google Cloud revenue more than doubled, indicating that enterprise clients are actively adopting the technology. However, investors, concerned about significant expenditures and the slowing growth of the dominant search engine, reduced the company's stock value by 10% after trading closed on Wednesday.

Dave Wagner, portfolio manager at Aptus Capital Advisors, noted that 'in a market worried that AI will displace the franchise, investors need to see acceleration in search, not deceleration.' He added that 'Cloud supports the history, but until search demonstrates that AI is an additive element, shares are likely to trade lower.'

According to LSEG data, the company reported a negative free cash flow of $9.1 billion, which was below analysts' expectations that projected near break-even. This marks the second consecutive quarter for Alphabet with negative free cash flow. In the previous quarter, aggressive spending on cloud capacity and AI development led to a burn of $5.9 billion, the first negative figure in 22 years for a public company.

Total revenue for the third quarter reached $127.6 billion, falling short of the analyst estimate of $128.2 billion. Although Google Cloud exceeded expectations, this was not enough to offset the $1.4 billion shortfall in the advertising sector; ad revenue reached $82 billion against an expected $83.4 billion. Net income available to Alphabet shareholders dropped by 25% to $26.17 billion, and adjusted earnings per share came in at $2.67, missing Wall Street's target of $3.02.

AI Mode

Chief Business Officer Philipp Schindler explained the stronger-than-expected slowdown in search advertising growth due to the broader use of the AI mode—the search engine's chat interface, which the company is still experimenting with for advertising. He informed analysts during the conference call that Google will begin implementing advertising in the AI mode in international markets.

CFO Anat Ashkenazi informed analysts that Alphabet still plans to significantly increase capital expenditures in 2027—from the current year's forecast of $195 billion to $205 billion. The executives did not provide a figure for next year, but investors expressed concern over such a large jump in spending.

Jeremy Salzburg, Senior Investment Analyst at Check Capital, stated that 'a capital expenditure figure of $300 billion could potentially scare the market, as free cash flow is now finally negative.' Ashkenazi also warned of potential short-term margin pressure due to the use of third-party cloud capacity.

These results reflect enormous expenses in building AI infrastructure, causing investors to worry both about potential overspending and the risk of falling behind competitors due to insufficient investment.

In contrast, competitor Microsoft, which also released its results on Wednesday, saw its shares rise by more than 5%, supported by positive free cash flow that exceeded analyst forecasts.

Alphabet attempted to compensate for the unsatisfactory advertising figures by highlighting its cloud computing division, notably by having Thomas Kurian, CEO of Google Cloud, speak on the earnings call for the first time.

Google Cloud revenue more than doubled to $30.71 billion in the third quarter ending September 30th, beating forecasts. Kurian noted that sales of Google's Tensor Processing Units (TPUs), which compete with Nvidia GPUs, have become a significant contribution to the business. In the previous quarter, Alphabet began recognizing revenue from direct sales of what it calls 'TPU systems,' which include chips and associated hardware and software.

However, Ashkenazi cautioned that increasing TPU inventory will put additional pressure on free cash flow. She reported that Google Cloud's order backlog grew to $561 billion, and Alphabet expects to recognize over 55% of it as revenue within the next 24 months.

Google Cloud, the third-largest cloud service provider globally after Amazon Web Services and Microsoft Azure, has expanded its enterprise commitments through multi-year agreements for cloud technology and AI infrastructure in cybersecurity, retail, and financial services.

It is projected that Alphabet, Microsoft, Amazon, and Meta combined will spend over $700 billion in 2026 to expand their data centers for AI, chips, and high-performance networking. Since cutting-edge chips and data center capacity are scarce, Alphabet's investments in proprietary TPUs alongside third-party GPUs have helped expand its capacity.

Gemini 4 Carbon

Earlier on Wednesday, Google attempted to preempt the earnings release and refute the notion that it is losing the AI race by introducing its new Gemini 4 Carbon model. CEO Sundar Pichai stated during the call that this means the company is once again 'at the forefront.'

Following this, Bernstein analysts asked in their note: 'Has Google rushed the deployment of Gemini 4 models to compensate for softer financial results? And are the models ready for commercial implementation?' - Akesh Sriram and Kenrick Kai.

Popular