Baidu has released its cutting-edge reasoning model ERNIE X1 alongside the multimodal ERNIE 4.5 on March 16, 2025, directly challenging competitors like DeepSeek and OpenAI. These new models combine exceptional performance with disruptive pricing, offering ERNIE X1 at half the cost of DeepSeek R1 while matching its capabilities and positioning ERNIE 4.5 at just 1% of GPT-4.5’s price while claiming superior results.
Key Takeaways
- ERNIE X1 delivers comparable performance to DeepSeek R1 at 50% of the cost
- Baidu will make ERNIE Bot free for individual users starting April 1, 2025
- ERNIE 4.5 claims to outperform GPT-4.5 while costing 99% less
- Baidu plans to open-source the ERNIE 4.5 series starting June 30, 2025
- The models feature technical innovations like Progressive Reinforcement Learning and FlashMask Dynamic Attention Masking
Disruptive Pricing Strategy Shakes Up AI Market
Baidu’s aggressive pricing approach for its new models represents a major market disruption. ERNIE X1’s API pricing starts at just RMB 0.002 ($0.28 per million tokens) for input and RMB 0.008 ($1.10 per million tokens) for output. This makes it 50% cheaper than DeepSeek R1 while offering comparable performance in reasoning tasks.
Even more surprising is ERNIE 4.5’s pricing at RMB 0.004 ($0.55 per million) for input and RMB 0.016 ($2.20 per million) for output, making it 99% less expensive than OpenAI’s GPT-4.5 while claiming to outperform it. This pricing strategy is forcing competitors to reconsider their cost structures as the high-performance AI model landscape becomes increasingly competitive.
Technical Innovations Powering ERNIE X1
ERNIE X1 isn’t just about competitive pricing; it introduces significant technical advancements. The model employs Progressive Reinforcement Learning and End-to-End Training to integrate reasoning chains effectively. This approach helps the model think step-by-step through complex problems, improving its logical reasoning capabilities.
Another innovation is FlashMask Dynamic Attention Masking, which enhances computational efficiency by optimizing how the model focuses on relevant information. The Heterogeneous Multimodal Mixture-of-Experts architecture enables efficient processing of diverse data types, allowing the model to handle text, images, and other inputs seamlessly.
ERNIE X1 also excels in tool use, supporting AI image generation, academic search, and business analytics. These capabilities make it particularly strong in Chinese knowledge Q&A, logical reasoning, and practical applications compared to other AI models in the market.
Performance Benchmarks Against Competitors
According to Baidu’s benchmarks, ERNIE X1 matches DeepSeek R1’s reasoning capabilities while costing half as much. Meanwhile, ERNIE 4.5 supposedly outperforms GPT-4.5 in text generation and multimodal tasks despite its dramatically lower price point.
The company claims a 30% reduction in hallucinations compared to previous models, along with improved code-writing accuracy and interpretation. ERNIE 4.5’s ability to process video, audio, and images with spatiotemporal compression gives it versatile multimodal capabilities, something Baidu hopes will differentiate it in an increasingly crowded market.
As noted by Bernstein Research, DeepSeek’s previous cost advantages have been effectively neutralized by Baidu’s strategy. Early adopters have praised ERNIE X1’s output quality and tool integration, with Vin Foo from ZeroLaunch highlighting its superior usability compared to competing models.
Strategic Ecosystem Integration and Accessibility
Baidu isn’t just launching these models in isolation; it’s integrating them across its ecosystem. ERNIE X1 will be incorporated into Baidu Search, the Wenxiaoyan app, and various enterprise tools, creating a cohesive AI experience for users.
In a bold move, Baidu announced it will open-source the Ernie 4.5 series starting June 30, 2025. This approach follows the trend of open AI model development that has helped accelerate innovation in the field.
The company’s free ERNIE Bot release targets consumer adoption amid rising competition from ByteDance and Moonshot AI. Baidu has also made substantial infrastructure investments, including a $2.1 billion acquisition of YY Live streaming platform to bolster its cloud-AI infrastructure. This has contributed to a 26% growth in cloud revenue in Q4 2024, driven by increasing demand for generative AI.
The Future of Baidu’s AI Strategy
Looking ahead, Baidu plans to expand its model capabilities through regular updates. The company is focusing on enterprise AI solutions across industries while continuing to invest in AI infrastructure and research.
Global market expansion is also on the horizon, with plans for localized model variations to serve different regions. Baidu’s commitment to an open-source ecosystem aims to accelerate AI development and foster a community of developers around its technology.
With these strategic moves, Baidu is positioning itself as a major player in the global AI landscape, challenging established competitors with a combination of technical innovation, aggressive pricing, and ecosystem integration.
Sources
techcrunch.com – Baidu launches two new versions of its AI model ERNIE
prnewswire.com – Baidu Unveils ERNIE 4.5 and Reasoning Model ERNIE X1
opentools.ai – Baidu Unleashes ERNIE X1 AI to Challenge DeepSeek in the Reasoning Race