TikTok’s parent company, ByteDance, has launched a new version of its popular AI app, Doubao, which is the leading consumer-facing AI tool in China. The updated version, called Doubao 1.5 Pro, brings enhancements in performance while also being more cost-efficient to run.
Key improvements in Doubao 1.5 Pro
The new Doubao 1.5 Pro uses a more efficient approach to training its AI model, which ByteDance says helps balance the system’s performance with lower costs. This is achieved through a design that combines both training and real-time use of the model, optimising it for better results while keeping the infrastructure costs low. The system is built to support a variety of chips, including lower-end options.
ByteDance’s Doubao 1.5 Pro is a cost-effective alternative to global AI models, offering a significant price advantage over competitors. At just $0.022 per million cached input tokens, $0.11 per million input tokens, and $0.275 per million output tokens, Doubao is five times cheaper than DeepSeek’s model and more than 200 times less expensive than OpenAI’s GPT-4o. Doubao also boasts a 32k + 256k context and utilises a sparse Mixture of Experts (MoE) architecture. The new pricing and technological enhancements strengthen Doubao’s position as a cost-efficient alternative to other advanced AI systems.
Doubao outperforms global competitors in tests
Doubao 1.5 Pro has already shown impressive results in performance tests. It outshone both domestic and global AI models from companies like OpenAI, Google, and Amazon’s Anthropic in certain areas such as language understanding, math, coding, and visual reasoning. In fact, it ranked as the best in half of the 14 benchmarks that tested its abilities.
ByteDance’s efficient AI strategy amid global chip shortages
ByteDance’s focus on efficiency is in line with trends seen in other tech companies, especially those in China, which are facing limitations on advanced chip access due to US export restrictions. By focusing on resource-efficient models, ByteDance is working to keep up with global AI leaders despite these challenges.
Competition from other AI companies
ByteDance’s Doubao 1.5 Pro follows the success of the AI company DeepSeek, which surprised the global tech community by introducing its own efficient AI model last year. DeepSeek’s model used far fewer resources than Meta’s Llama 3.1, offering similar performance at a much lower cost.
While ByteDance hasn’t revealed how much it spent on training Doubao 1.5 Pro or the specific chips used, it continues to innovate in China’s AI space, making significant strides despite external pressures.