Alibaba

ByteDance

Huawei

Meituan

Tencent

Xiaomi

Most Followed

KrASIA Connection

100 AI Creators

China's Auto Sector

Inside China's Startups

Venture Roundup

Our Events

View All

Collections

News

Features

Insights

Ecosystem

Press Releases

Anker

DeepSeek

Li Auto

MiniMax

Pop Mart

Xpeng

On the Radar

Events

Latest

About

Consumer

Robotics

Fintech

Collaborate

… hybrids over three times the range of traditional gasoline cars, driving the company’s overall 41.3% sales growth.
But pure EVs remain the ultimate goal of electrification. That’s why BYD is doubling down on its super e-platform.
The Han L and Tang L will be the first to feature this platform. These flagship models, part of BYD’s Dynasty series, combine a new design language with the super e-platform, the DiPilot 300 smart driving system, and the DiSus-C chassis system.
The Han L targets competitors like <a href="https://kr-asia.com/chinas-xiaomi-jumps-into-ev-market-battle-with-gusto">Xiaomi’s SU7</a>, while the Tang L positions itself against Li Auto’s L6 and future family-focused EVs like the i6.
BYD’s biggest challenge remains selling high-end EVs in a market where plug-in hybrids dominate.
At present, ultrafast charging seems to be the only way to make pure EVs more practical for everyday use.
But when will this technology reach BYD’s mass market models like the Qin and Song series? The company has not made an official statement, but engineers at the event told 36Kr that “it’s only a matter of time.”
BYD’s DiPilot 300 system took one and a half years to move from premium models to mainstream vehicles. But that was a software-driven upgrade. Megawatt charging is hardware-dependent, meaning adoption will likely take longer—perhaps two to three years.
Beyond battery upgrades, entire vehicle systems—from power control to drivetrain components—must be optimized for high-voltage operation, adding costs in the thousands of dollars per unit.
BYD’s edge lies in its vertical integration. With in-house kilovolt architecture and high-voltage component manufacturing, it can reduce costs more effectively than competitors.
Still, mass adoption will require a full model refresh and greater economies of scale. With the Han L, Tang L, and upcoming Song L and Qin L, BYD is gradually phasing out older models to accelerate this transition.
But without a robust charging network, even the fastest-charging EVs won’t reach their full potential, making the next two to three years critical.
By then, megawatt charging could be mainstream, while solid-state batteries may still be years from commercialization.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3211032618648710">article</a> was written by Han Yongchang for 36Kr.

BYD’s new EVs could charge faster than your smartphone

East Asia

China

Transportation

Electric Vehicles

Deep Tech

Batteries

Mobility

Energy

… advisory.</li>
<li aria-level="1">No requirement for a physical office, with access to shared or virtual office spaces at no additional cost.</li>
<li aria-level="1">Dedicated Chinese-speaking support staff, offering tailored assistance to Chinese enterprises.</li>
</ul>
For companies eyeing the Middle East market, the UAE remains a strategic, business-friendly destination with significant advantages in taxation, infrastructure, and market accessibility. By leveraging the right strategies and resources, businesses can establish a strong foothold in this dynamic region.

Expanding to the UAE? Here are 5 things businesses need to know

Middle East

On March 16, <a href="/company/baidu">Baidu</a> officially launched <a href="https://kr-asia.com/baidu-slashes-ai-costs-with-ernie-4-5-and-x1-taking-on-openai-and-deepseek">Ernie X1 and Ernie 4.5</a>, two advanced large models now available for free on the Ernie Bot website and poised to become the next key drivers of artificial intelligence adoption.
From <a href="/company/openai">OpenAI</a> to Grok, Google’s Gemini, and beyond, AI’s major players are constantly shuffling. The one constant? AI applications are still in the early stages of commercialization.
Baidu’s latest rollout signals its intention to prepare for the anticipated AI app boom in 2025, facilitated by multimodal AI.
Ernie 4.5 introduces a breakthrough in native multimodal learning, improving comprehension across text, images, and logic-based tasks. By integrating joint modeling across multiple modalities, it enhances textual reasoning and logical inference. Benchmark results suggest that Ernie 4.5 outperforms OpenAI’s GPT-4.5—while offering API access at just 1% of GPT-4.5’s cost.
Meanwhile, Ernie X1 is an advanced reasoning model positioned against <a href="https://kr-asia.com/a-new-ai-disruptor-businesses-rush-to-integrate-deepseek-r1">DeepSeek-R1</a>. Baidu claims X1 offers comparable capabilities in multiple benchmarks, with pricing at only half that of DeepSeek-R1. Unlike traditional models focused on content generation, X1 is engineered for decision-making, pushing AI beyond a response tool to a more interactive and analytical agent.
For instance, in narrative generation tests, X1 can craft intricate murder mystery plots from a simple background prompt. It also excels at mimicking the sharp, opinionated tone prevalent in Chinese social media, making it a potential tool for content creators seeking more engaging, localized AI-generated responses.
Since launching Ernie 4.0, Baidu has rapidly built an AI ecosystem, integrating AI-powered search, maps, cloud storage, and document services. Back in 2021, the company transitioned from its strategy of integrating AI and cloud to prioritizing AI-first architecture.
The free release of Ernie 4.5 and X1 reflects cost reductions driven by technological advancements. This move also signals three key industry trends:
<ol>
<li aria-level="1">Baidu has a rare advantage as one of the few AI firms with a four-layer architecture approach, encompassing foundational research, framework, model, and application. Its deep expertise in AI chips and infrastructure supports long-term commercialization efforts.</li>
<li aria-level="1">China’s tech giants are not just competing on models but also model-as-a-service (MaaS) platforms. Enterprises and developers can now access Ernie 4.5 APIs via Qianfan, while Ernie X1 will be added soon.</li>
<li aria-level="1">AI adoption in China has reached a tipping point as businesses grow eager to embrace new AI technologies. Baidu, as a leader in the space, can lower entry barriers for developers and enterprises to accelerate AI industrialization.</li>
</ol>
A recent evaluation demonstrated Ernie 4.5's notable ability to reduce AI hallucinations. This improvement stems from FlashMask dynamic attention masking for improved accuracy, a heterogeneous multimodal mixture-of-experts (MoE) for optimized reasoning, and a self-feedback enhanced post-training process.
In testing, when presented with a single movie screenshot, Ernie 4.5 was able to accurately identify the film.
Its chart analysis and reasoning capabilities also highlight its enhanced multimodal generation abilities.
The industry’s interest in multimodal large models is no secret.
By jointly modeling images, video, and text, AI can unify semantic understanding and overcome the fragmented information issues of conventional chatbot-style models.
For example, model context protocol (MCP) has drawn significant attention in “vibe design” on Blender, where AI facilitates iterative design refinements through dialogue-based interactions, making AI more of a controlled creative tool rather than a passive responder.
Similarly, Gemini’s multimodal capabilities in Google AI Studio—handling multi-turn conversations, image generation, and real-time editing—have positioned it as a go-to AI for developers.
Baidu's Ernie 4.5 is ostensibly China’s first native multimodal model. Meanwhile, Ernie X1, also multimodal, particularly excels in Chinese Q&amp;A, literature generation, document drafting, logical reasoning, complex calculations, and tool usage.
X1's strengths are rooted in progressive reinforcement learning, improving creativity, search, tool usage, and logical inference across multiple domains. Additionally, its end-to-end training based on reasoning and action chains has enhanced its ability to perform deep search and tool calls, which are areas where many AI models still struggle.
On the cost side, PaddlePaddle and Ernie have implemented optimizations in model compression, inference engines, and system architecture, achieving deep compression and accelerated inference. This is why X1’s costs are only half those of DeepSeek-R1.
As the AI industry enters the era of deep reasoning, falling costs have eliminated one of the biggest barriers to adoption.
The two major challenges that enterprises face in AI deployment are high technical barriers and unsustainable costs.
Small and midsized businesses often struggle with AI costs, while larger enterprises, despite having technical teams, face high training expenses and complex adaptation challenges.
For years, the feasibility of AI adoption was unclear, leading to inefficient investments. But as AI models continue to improve, companies across industries are now fully embracing AI-driven transformation.
Baidu’s strategy of lowering costs and raising accessibility with Ernie 4.5 and X1 is aimed at addressing these pain points.
whereRelatedArticle
Baidu’s cloud AI business has steadily grown over the past year, proving that more enterprises are leveraging MaaS platforms and foundational AI models to build applications without excessive upfront investment.
In March 2023, Baidu was the first Chinese firm to declare its commitment to rebuilding all of its products with AI. Since then, the company has invested heavily in next-gen foundational models, culminating in today's native multimodal Ernie models.
2025 could be the breakout year for enterprise AI adoption, where precision and accuracy will be paramount. Baidu, with Ernie 4.5 and X1, could yet be the one leading the charge.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3210023586514052">article</a> was written by Xiao Xi for 36Kr.

Baidu lowers the cost of AI with Ernie 4.5 and X1

Baidu

Artificial Intelligence

Foundation and Large Models

<a href="http://www.dbaspace.com/about.html">Deep Blue Aerospace</a>, a company specializing in liquid-fueled recoverable and reusable launch vehicle technology since its founding in 2016, has secured nearly RMB 500 million (USD 70 million) in a new financing round led by Taian Yuanwang New Energy Industry Investment Fund. Its core business spans R&amp;D, manufacturing, testing, and commercial launch services. By developing reusable rockets designed to deliver payloads into predetermined orbits with efficiency and precision, the company aims to offer high-quality and cost-effective space transportation solutions.
A recent power system test firing—a 200-second run documented by the company—showcased critical system reliability ahead of full-scale launch operations. This progress coincides with SpaceX’s milestone on October 13, 2024, when it conducted the fifth test flight of its Starship, successfully recovering the booster in a soft landing. That achievement, SpaceX’s first complete booster recovery since the vehicle’s inception, underscores the cost-saving potential of reusable launch technologies—a field in which China is making significant strides. Currently, no other Chinese firm has completed a launch mission with a liquid-fueled recoverable rocket.
Deep Blue Aerospace’s Nebula-I is China’s first recoverable liquid-fueled launch vehicle. The rocket boasts a payload capacity of two tons, a total thrust of 180 tons, and the capability for rapid turnaround through quick maintenance and relaunch. It is engineered for both orbital insertion and suborbital recovery, aligning with the growing demand for low-cost, high-reliability, and high-frequency launches.
Building on this foundation, the company is developing Nebula-II.
The Nebula-II’s first stage is powered by 11 custom Thunder-RS engines operating in parallel. This configuration is designed to lift more than 25 tons into low Earth orbit with over 1,000 tons of total thrust, supporting large-scale cluster launch demands, such as those required for satellite constellations like Starlink.
The Thunder-RS engine, which incorporates needle valve injection technology similar to that used by SpaceX, can generate 130 tons of thrust at ground level and approximately 150 tons in vacuum. With a throttle range extending from 30% to 110% and a reuse capability exceeding ten flights, the engine is set to play a pivotal role in reducing launch costs. It is manufactured using a 3D printing-integrated process that minimizes part count and shortens production cycles.
Deep Blue Aerospace has recently initiated full-system tests of the Thunder-RS engine. The Nebula-I has already passed key evaluations, including engine tests and control system verifications, affirming the maturity and reliability of its core technologies. Additionally, a ten-meter-level vertical flight recovery test has further validated the company’s recovery systems.
Since launching its Series B financing in May 2024, Deep Blue Aerospace has completed multiple rounds, bringing its cumulative funding to over RMB 1 billion (USD 140 million) and pushing its current valuation to approximately RMB 5 billion (USD 700 million).
On the industrial front, Deep Blue Aerospace has established an expansive liquid rocket engine test site in Jinan, Shandong—the largest of its kind in China by land area. By clustering the manufacturing of core engine components around the test site and initiating tests immediately after production, the company has significantly shortened its supply chain. Furthermore, it is accelerating an expansion strategy that leverages local industrial advantages to build a vertically integrated ecosystem covering propulsion systems, vehicle manufacturing, launch recovery, and reuse optimization.
Deep Blue Aerospace plans to conduct a new round of high-altitude recovery tests for the Nebula-I, with a launch slated for midyear to validate both orbital insertion and vertical recovery. The maiden flight of the Nebula-II is set for 2026, while future R&amp;D efforts will focus on developing higher-thrust engines and larger rockets—crucial for the expansion of satellite internet and deeper space exploration.
​​​​KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3192980080697730">article</a> was written by Huang Nan for 36Kr.

Deep Blue Aerospace raises fresh capital as China ramps up reusable rocket development

Venture Capital

Space Tech

Investment

… challenging the long-held perception of a relaxed tech work environment.
As the global economy slows and uncertainty looms, the 996 mentality has begun to spread overseas, just as Chinese firms experiment with work-life balance. But whether forced clockouts can become a sustainable long-term policy remains an open question.
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3202609303265794">article</a> was written by Hu Yiting for 36Kr.

“996” no more? DJI and Midea are making employees leave work on time

… lighting, half-leather, half-fabric seats, and the removal of heated seats and a heated steering wheel—trimming USD 4,000 off the original price.
However, applying the same approach in China is risky. Domestic EV brands have redefined expectations for in-car luxury, offering large entertainment displays and advanced driver assistance systems. At times, Tesla’s minimalist interior has drawn flak for being too barebones in comparison.
If Tesla relies solely on price cuts, it faces an uphill battle.
Tesla has seen the benefits of aggressive pricing before. In late 2024, the company introduced a zero-interest, five-year financing plan in China, saving buyers RMB 15,000 (USD 2,100) in interest on a RMB 100,000 (USD 14,000) loan. The move temporarily boosted Model Y sales, but domestic competitors quickly countered with similar or better financing options.
The lower-cost Model Y may attract price-sensitive buyers, but Tesla cannot rely solely on discounts. Unlike competitors that continuously upgrade models, Tesla’s supply chain is less flexible, making frequent updates more challenging.
At best, a cheaper Model Y might provide short-term relief, but it does not resolve Tesla’s deeper struggles in China.
Tesla currently offers four models in China: the Model 3 and Model Y, priced between RMB 200,000–300,000 (USD 28,000–42,000), and the Model S and Model X, which start at RMB 600,000 (USD 84,000). This year, the company is expected to introduce two lower-cost versions of the Model 3 and Model Y.
Meanwhile, Tesla’s China team is working to bring the <a href="https://kr-asia.com/the-cybertruck-took-four-years-to-deliver-but-it-could-mark-teslas-leap-into-a-new-technological-era">Cybertruck</a> to market, though its prospects for mass adoption remain limited. Unlike in the US, pickup trucks have minimal appeal among Chinese consumers.
<figure id="attachment_146921" aria-describedby="caption-attachment-146921" style="width: 1024px" class="wp-caption alignnone"><img loading="lazy" decoding="async" class="size-large wp-image-146921" src="https://console.kr-asia.com/wp-content/uploads/2023/12/Tesla-Cybertruck-3-1024x683.jpg" alt="Image of the Cybertruck." width="1024" height="683" srcset="https://console.kr-asia.com/wp-content/uploads/2023/12/Tesla-Cybertruck-3-1024x683.jpg 1024w, https://console.kr-asia.com/wp-content/uploads/2023/12/Tesla-Cybertruck-3-300x200.jpg 300w, https://console.kr-asia.com/wp-content/uploads/2023/12/Tesla-Cybertruck-3-768x512.jpg 768w, https://console.kr-asia.com/wp-content/uploads/2023/12/Tesla-Cybertruck-3-1536x1024.jpg 1536w, https://console.kr-asia.com/wp-content/uploads/2023/12/Tesla-Cybertruck-3.jpg 1920w" sizes="(max-width: 1024px) 100vw, 1024px" /><figcaption id="caption-attachment-146921" class="wp-caption-text">Image of Tesla’s Cybertruck. Image source: Tesla.</figcaption></figure>
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3202436263830017">article</a> was written by Xu Caiyu for 36Kr.

Facing a tough test in China, Tesla’s next move is a budget Model Y

Tesla

Excitement is mingling with skepticism as Chinese tech companies that hope to be seen as "the next <a href="/company/deepseek">DeepSeek</a>" roll out one artificial intelligence breakthrough after another.
In one of the most recent examples, early access codes for Manus AI are reportedly changing hands for thousands of dollars amid a frenzy of online interest in the app.
The company, also called Manus AI, released a demo video on March 5 claiming the world's "first general AI agent" can do everything from screening resumes to researching real estate and analyzing stocks.
"This isn't just another chatbot or workflow. It's a truly autonomous agent that bridges the gap between conception and execution. While other AI stops at generating ideas, Manus delivers results," Ji Yichao, co-founder and chief scientist of Manus AI, said in the video.
"We see it as the next paradigm of human-machine collaboration, and potentially a glimpse into AGI," Ji added, referring to the yet-unrealized concept of artificial general intelligence (AGI) in which AI achieves human-level cognitive abilities.
The company also claims Manus surpassed ChatGPT developer OpenAI's models across all three difficulty levels in the GAIA benchmark, which assesses general AI assistants on their real-world problem-solving capabilities.
"This wouldn't be possible without the amazing open-source community. … Manus operates as a multi-agent system powered by several distinct models, so later this year, we're going to open-source some of these models, specifically post-trained for Manus, inviting everyone to explore this agentic future together," Ji said.
The app is not yet publicly available as it requires an invitation code to use, but it quickly took the Chinese internet by storm after a number of tech influencers who were given early access shared their impressions of the agent. The official website crashed due to a surge in traffic, and invitation codes became so scarce that some were reportedly being resold online for as much as RMB 50,000 (USD 7,000).
After the launch, many Chinese media outlets began referring to Manus AI as "the next DeepSeek."
Compared with the frenzy in China, the overseas reception to the app has been lukewarm, even though both its website and the demo video are in fluent English. Its video on YouTube has generated around 28,000 views as of Friday morning, while its X account is tentatively suspended over "rule violation."
However, the sudden popularity of the app has also triggered skepticism as only a handful of users have experienced it firsthand. Some critics argue that the so-called DeepSeek moment for AI applications is just little more than a "marketing frenzy" based on hype surrounding open-source models.
"The core technology of Manus is its ability to automatically deploy different tools and models, something other companies haven't achieved," said an AI engineer not related to the company and who asked not to be named. "Manus claims its accuracy is very high, while the industry standard for accuracy is actually quite low."
The engineer expressed skepticism, however, about the uniqueness of Manus' achievement and compared it to foreign competitors like Anthropic's Claude Computer Use. He also pointed to the fevered atmosphere around the country's AI industry.
"It feels like the AI industry right now is all about China's rise, so whenever something AI-related comes up, people immediately start hyping it up. I'm worried this might disrupt normal development, though of course, appropriate encouragement is also important."
Under mounting public pressure, Manus AI partner Zhang Tao said on March 6 that the company "has never allocated any budget for marketing" and that the invite-only system was implemented purely due to "server capacity limitations."
Manus AI is not the only company looking to follow in the footsteps of startup DeepSeek, which took the tech world by storm in late January when it unveiled low-cost inference models based on open-source technology.
On March 6, Alibaba Group released its own open-source AI reasoning model, dubbed QwQ-32B, saying it outperforms DeepSeek's R1 in areas such as mathematics, coding, and general problem-solving, despite its relatively modest 32 billion parameters. A larger number of parameters generally means a more capable technology, but it also requires more computing power. In comparison, DeepSeek's R1 boasts 671 billion parameters.
whereRelatedArticle
Whether claims like these turn out to be hype or reality, Beijing is throwing its weight behind the sector, too, in an apparent hope that advances in AI and other technologies can boost the economy amid curbs from the West. On March 5, Chinese premier Li Qiang vowed in his annual government work report that authorities will continue to support research and development in core technology for key fields and increase funding for fostering futuristic technologies such as embodied AI this year.
Investors, seizing on Beijing's promises of policy support and announcements by companies like Manus and Alibaba, have piled into AI-related shares, pushing the Hang Seng Tech Index to its highest level since late 2021 on March 7.
This article first appeared on <a href="https://asia.nikkei.com/Business/Technology/The-next-DeepSeek-Startup-Manus-becomes-China-s-latest-AI-darling">Nikkei Asia</a>. It has been republished here as part of 36Kr’s ongoing partnership with <a href="https://www.kr-asia.com/nikkei-teams-up-with-36kr-in-asia-tech-news-coverage/">Nikkei</a>.

The next DeepSeek? Startup Manus becomes China's latest AI darling

… simulation, Tripo aims to simplify the creation of immersive environments, bridging the gap between static assets and interactive worlds.
And the impact of AI extends far beyond digital entertainment.
In a designer toy factory in Dongguan, China, handdrawn sketches are now converted directly into 3D-printable files using Tripo’s API. A process that once took two months now takes just three days—with AI even optimizing designs for structural integrity.
Meanwhile, fashion design platform Look has been using AI since 2023 to transform sketches into 3D clothing models. The system automatically refines fabric textures, adjusts materials, and adapts how outfits drape on different body types.
<figure id="attachment_152232" aria-describedby="caption-attachment-152232" style="width: 720px" class="wp-caption aligncenter"><img loading="lazy" decoding="async" class="wp-image-152232" src="https://console.kr-asia.com/wp-content/uploads/2025/03/Tripo-AI-3D-Model-1-1024x512.jpg" alt="Photo of a 3D model produced using a design generated by Tripo." width="720" height="360" srcset="https://console.kr-asia.com/wp-content/uploads/2025/03/Tripo-AI-3D-Model-1-1024x512.jpg 1024w, https://console.kr-asia.com/wp-content/uploads/2025/03/Tripo-AI-3D-Model-1-300x150.jpg 300w, https://console.kr-asia.com/wp-content/uploads/2025/03/Tripo-AI-3D-Model-1-768x384.jpg 768w, https://console.kr-asia.com/wp-content/uploads/2025/03/Tripo-AI-3D-Model-1.jpg 1080w" sizes="(max-width: 720px) 100vw, 720px" /><figcaption id="caption-attachment-152232" class="wp-caption-text">Photo of a 3D model produced using a design generated by Tripo. Photo source: 36Kr.</figcaption></figure>
<h3>New frontier for digital creation</h3>
From Where Winds Meet’s AI-driven world-building to indie developers crafting entire games in hours, AI is reshaping the creative landscape. But as these tools become more powerful, they also raise new questions about the relationship between imagination and reality—especially now that nearly anyone can build hyper-realistic virtual worlds with minimal effort.
This shift is both technological and cultural. AI is democratizing content creation, lowering the barrier for individuals to bring their ideas to life. And the effects are already rippling across industries:
<ul>
<li aria-level="1">Game development: Over 200 games on Steam now integrate AI-generated content.</li>
<li aria-level="1">3D printing: On Taobao, 45% of new 3D print orders are for game-related models.</li>
<li aria-level="1">Film production: Studios are starting to source AI-assisted player creations for digital assets.</li>
</ul>
As the line between virtual and real continues to blur, the question isn’t just what AI can create—it’s what we choose to do with it. The future of digital creativity won’t be defined by AI alone, but by the people who wield it.
Or, as Tripo’s founder Song Yachen put it:
<blockquote>"We’re not just teaching AI to generate models—we’re unlocking the imagination of an entire generation."</blockquote>
KrASIA Connection features translated and adapted content that was originally published by 36Kr. This <a href="https://36kr.com/p/3192960825090690">article</a> was written by Liu Shiwu for 36Kr.

AI-powered tools like Tripo could make game development faster, smarter, and more creative

Gaming

Creative

<a href="/company/baidu">Baidu</a> has launched Ernie 4.5, its latest multimodal foundation model, and Ernie X1, a deep-thinking reasoning model, making both available for free via Ernie Bot.
The move underscores Baidu’s push to undercut <a href="/company/openai">OpenAI</a> and <a href="/company/deepseek">DeepSeek</a> with low-cost, high-performance AI alternatives. Ernie 4.5, a native multimodal model, processes text, images, audio, and video, with improved logical reasoning, memory, and reduced hallucinations. Baidu claims it outperforms GPT-4.5 while costing just 1% of its price.
Ernie X1, designed for complex reasoning and tool use, supports advanced search, document Q&amp;A, image generation, and code interpretation. Baidu has positioned it as a cheaper alternative to DeepSeek R1, offering similar performance at half the price.
Baidu’s latest models mark a shift toward high-performance AI at significantly lower costs. Ernie 4.5’s token pricing starts at RMB 0.004 (USD 0.00056) per 1,000 tokens for input and RMB 0.016 (USD 0.0022) for output, well below OpenAI’s GPT-4.5. Ernie X1 costs even less, starting at RMB 0.002 (USD 0.00028) for input and RMB 0.008 (USD 0.0011) for output.
whereRelatedArticle
Beyond pricing, Baidu has made technical refinements in both models.
Ernie 4.5 integrates FlashMask dynamic attention masking for improved accuracy, a heterogeneous multimodal mixture-of-experts (MoE) for optimized reasoning, and a self-feedback enhanced post-training process to reduce hallucinations.
Ernie X1, meanwhile, incorporates a progressive reinforcement learning method and an end-to-end training approach, improving structured reasoning and problem-solving.
The launch of Ernie 4.5 and X1 sets the stage for Ernie Bot 5.0, which Baidu plans to release by mid-2025. The company has been iterating rapidly, bringing new models to market in quick succession. Ernie Bot made its <a href="https://kr-asia.com/baidus-ernie-bot-among-eight-chinese-llm-products-approved-for-public-launch">public debut</a> in August 2023, making Baidu one of the earliest Chinese firms to introduce an AI chatbot in response to OpenAI’s ChatGPT. <a href="https://kr-asia.com/baidu-claims-its-latest-ai-model-ernie-4-0-is-on-par-with-openais-gpt-4">Ernie 4.0</a> arrived in October 2023, claiming parity with GPT-4, followed by <a href="https://kr-asia.com/baidu-deepens-foothold-in-chinese-ai-market-with-ernie-4-0-upgrade">Ernie 4.0 Turbo</a> in June 2024, which reportedly improved response speeds and expanded API availability.

Baidu slashes AI costs with Ernie 4.5 and X1, taking on OpenAI and DeepSeek