Doubao Pro (1.5 Pro)

ByteDance’s Doubao 1.5 Pro (aka “Doubao Pro 1.5”) is a Mixture-of-Experts model designed to balance top-tier capability with low inference cost through an integrated training–inference approach. The team highlights a highly efficient sparse-MoE architecture and extensive RL/post-training that push reasoning, coding, and multilingual skills; the same model line also upgrades vision and speech, supporting document/image understanding and end-to-end speech interactions.

For scale and deployment, Doubao 1.5 Pro is offered in 32K and 256K context configurations and is accessible to developers through Volcengine (火山引擎) and partner gateways compatible with OpenAI-style APIs. Media and docs around the launch also note a 256K variant and generous output lengths.

Pricing is positioned as ultra-low, with third-party guides listing the 32K tier around ¥0.8 per million input tokens and ¥2 per million output tokens, and higher—yet still inexpensive—rates for 256K. (Exact rates can vary by account/region.)

Overall, Doubao 1.5 Pro is ByteDance’s value-performance play: a reasoning-centric, multimodal MoE model with large-context options, broad API availability, and unusually aggressive pricing for production workloads.