ByteDance Officially Launches DouBag Seed2.1 Large Model API: Pro Version Benchmark Surpasses GPT-5.5

According to Dynamic Beating Monitoring, ByteDance's Volcano Engine has officially launched the Doubao-Seed-2.1 series of large models and opened the API, with all models uniformly providing a 256k context window. The new series includes two deep thought models: Pro and Turbo, focusing on upgraded complex programming delivery, long-chain intelligent agent execution, and multi-modal visual capabilities.

In the benchmark tests disclosed by the company, the flagship Doubao-Seed-2.1-pro model has approached or surpassed GPT-5.5 in various code and development evaluations. In the SciCode scientific computing and NL2Repo-Bench library-level code generation tests, the model scored 59.8% and 47.0%, respectively, outperforming GPT-5.5, which scored 58.4% and 45.1%. In the software engineering evaluation SWE-Pro, the model achieved a score of 57.5%, higher than Gemini-3.1-Pro's 54.2% and close to GPT-5.5's 58.6%.

Regarding API pricing, the Pro version is priced at 6 yuan per million tokens for inference input and 30 yuan per million tokens for inference output. The Turbo version is priced at half the rate, at 3 yuan per million tokens for inference input and 15 yuan per million tokens for inference output.

The Volcano Engine has also simultaneously launched Seed-Evolving with weekly rolling upgrades. Seed-Evolving supports developers in obtaining the latest model capabilities iterated at least once a week through a unified Model ID, transitioning the large model delivery mode from a fixed version to continuous online evolution.