Baidu Open-Sources Ernie 4.5, Unveils New AI Models to Challenge Competitors with Lower Costs
Baidu Unleashes AI Power: Ernie 4.5 Goes Open-Source, Challenges Rivals with Cost-Effective Models
Baidu has significantly escalated the AI competition by open-sourcing its latest Ernie 4.5 model family and introducing new, highly cost-effective AI models, including Ernie X1. This strategic move aims to democratize advanced AI capabilities, challenging industry leaders like OpenAI and DeepSeek with competitive performance at substantially lower prices.
Baidu's Bold Open-Source Strategy
Baidu's decision to open-source its Ernie 4.5 model family under the Apache 2.0 license marks a pivotal shift in its AI strategy. Previously, Baidu kept its Ernie models proprietary. This move is seen as a direct response to the growing influence of open-source models, particularly from Chinese competitors like DeepSeek, and aligns Baidu with a global trend towards more accessible AI development.
- Ernie 4.5 Family: This comprehensive release includes ten variants, ranging from compact 0.3 billion parameter versions to sophisticated Mixture-of-Experts (MoE) models with up to 424 billion total parameters. These models are optimized for general-purpose language understanding and generation, supporting both textual and visual modalities.
- Developer Tools: Baidu has also made development toolkits like ERNIEKit and the FastDeploy deployment toolkit open source, further empowering developers to build powerful AI applications.
Introducing Ernie X1: A New Reasoning Powerhouse
Alongside the open-sourcing of Ernie 4.5, Baidu unveiled Ernie X1, its first dedicated reasoning model. Ernie X1 is designed for complex reasoning and tool use, excelling in tasks such as:
- Chinese knowledge Q&A
- Literary creation
- Manuscript writing
- Dialogue
- Logical reasoning
- Complex calculations
Baidu claims Ernie X1 offers performance comparable to DeepSeek R1 at half the cost, positioning it as a highly competitive alternative in the reasoning model landscape.
Unprecedented Cost Efficiency and Performance Claims
Baidu is aggressively undercutting the pricing of its competitors, making advanced AI more accessible. Key pricing details include:
- Ernie 4.5: 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. Baidu asserts Ernie 4.5 outperforms GPT-4.5 while costing only 1% of its price.
- Ernie X1: Even more cost-effective, starting at RMB 0.002 (USD 0.00028) for input and RMB 0.008 (USD 0.0011) for output.
These pricing strategies, combined with claims of matching or surpassing benchmarks set by models like DeepSeek R1 and OpenAI's GPT-4.5, signal Baidu's intent to disrupt the market by offering high-performance AI at significantly reduced costs.
Technical Innovations and Future Outlook
Baidu has implemented several technical refinements in its new models:
- Ernie 4.5: Features 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: Incorporates a progressive reinforcement learning method and an end-to-end training approach, enhancing structured reasoning and problem-solving.
Baidu plans to integrate both Ernie 4.5 and Ernie X1 into its product ecosystem, including Baidu Search. The company is also rapidly iterating, with plans to release Ernie Bot 5.0 by mid-2025, further solidifying its position in the global AI race.
Sources
- China's Baidu says its Kunlun chip cluster can train DeepSeek-like models, Reuters.
- Baidu slashes AI costs with Ernie 4.5 and X1, taking on OpenAI and DeepSeek, KrASIA.
- Baidu releases its first reasoning model ErnieX1 as AI competition ramps up, Cybernews.
- Baidu releases open source model family ERNIE 4.5, The Decoder.
- Baidu makes foundation model Ernie 4.5 open source, TechTarget.
