ByteDance Seed: Seed 1.6

ByteDance Seed: Seed 1.6 is a general-purpose text model from the ByteDance Seed team, released on 2025-12-23. It sits in the text category and is built around multimodal input handling combined with adaptive deep thinking, which means the model can take in more than just typed prompts and can adjust how much internal reasoning it applies to a request.

On the input side, Seed 1.6 accepts text, images and video, and it returns text. The context window is 262,144 tokens, and the maximum output is 32,768 tokens. That combination of a large context and multimodal input makes it suitable for tasks where the material to reason over is long or mixed in format, such as a lengthy document with embedded figures or a video that needs to be described and analysed alongside written instructions.

Practical use cases follow directly from those capabilities. You can send long text corpora for summarisation, question answering or extraction, and you can include images or video when the task depends on visual content, for example describing scenes, answering questions about a clip, or combining visual evidence with written context. The adaptive deep thinking behaviour is useful for problems that benefit from step-by-step reasoning, while simpler requests can be handled without that overhead. Structured outputs and tool calling also make it a reasonable fit for pipelines that need machine-readable results or access to external functions.

Through api14 you call the model with the id "bytedance-seed/seed-1.6" using the OpenAI-compatible API, so existing OpenAI SDKs and clients work with only a base URL and key change. The supported request parameters are frequency_penalty, include_reasoning, max_tokens, reasoning, response_format, stop, structured_outputs, temperature, tool_choice, tools and top_p. Reasoning-related controls let you request or expose the model's thinking, response_format and structured_outputs cover JSON and schema-constrained replies, and tools with tool_choice handle function calling.

Keep the output ceiling in mind: 32,768 tokens is the maximum response, so very large generations need to be split or streamed across calls. If your workload is text only and does not need visual input or explicit reasoning controls, a simpler text model may be a lighter fit, but for long-context, multimodal work with structured output and tool use, Seed 1.6 covers those requirements directly.