Volcano Engine launches Doubao 1.6 (Seed1.6) at its FORCE conference

Launch

On 11 June 2025 Volcano Engine, the cloud unit of ByteDance, released Doubao Large Model 1.6 at its FORCE conference.^[1] The model is also known as Seed1.6, the name used by ByteDance Seed.^[2]

Pricing

The model was priced by input length, with one rate that applied to thinking, multimodal and plain-text use. For inputs up to 32K, the price was 0.8 yuan per million input tokens and 8 yuan per million output tokens.^[1] Volcano Engine said its overall cost was about one third of Doubao 1.5 deep-thinking or DeepSeek R1.^[1]

A token is the unit a model reads and writes, so a price per million tokens is what a business pays for a large volume of text. Pricing by input length means a longer request is charged at a different level from a short one.

Architecture

ByteDance Seed described Seed1.6 as a mixture-of-experts model with 23 billion active parameters and 230 billion in total. Only part of the model is switched on for each input, which keeps the cost of running it below that of a dense model of the same total size. The model has a 256K context window.^[2]

Variants

The series includes Seed1.6-Thinking, which reasons before it answers, and an adaptive-thinking version that decides for each prompt whether to reason at all.^[2]

Exam results

Seed1.6-Thinking scored 683 on the humanities papers and 676 on the science papers of Shandong's 2025 gaokao, China's national college entrance examination.^[2]

Pricing that depends on input length is common in cloud services, because a longer request takes more resources to process. A reasoning model can also produce far more text before its final answer, so the output side of the bill can make up a large part of the cost.

Significance

The launch centred on two points: lower prices for reasoning-capable models and a model that decides when to think.