Doubao Seed family

ByteDance’s Doubao Seed text models, including Seed 2.1 Pro, Seed 2.1 Turbo, Seed Evolving and Seed Character. They are called through the OpenAI-compatible API, and all except Seed Character also support the Anthropic-native API, so they work in Claude Code. For Doubao’s image generation models, see Seedream (Doubao); for video models, see Seedance.

Common tasks

Models

For the full model list and live prices, see the model page . The model ID prefix is volcengine/ and can be omitted; the bytedance/ or doubao/ prefixes, or Volcano Engine Ark model names with dates, return 404.

These models are served by Volcano Engine (volcengine) and all support the Chat Completions and Responses APIs.

Examples

import os
from openai import OpenAI
 
client = OpenAI(base_url="https://api.ofox.ai/v1", api_key=os.environ["OFOX_API_KEY"])
 
resp = client.chat.completions.create(
    model="volcengine/doubao-seed-2.1-pro",
    messages=[{"role": "user", "content": "Explain idempotency in one sentence."}],
    extra_body={"thinking": {"type": "disabled"}},
)
print(resp.choices[0].message.content)

curl https://api.ofox.ai/v1/chat/completions \
  -H "Authorization: Bearer $OFOX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "volcengine/doubao-seed-2.1-pro",
    "messages": [{"role": "user", "content": "Explain idempotency in one sentence."}],
    "thinking": {"type": "disabled"}
  }'

Deep thinking

• thinking.type accepts enabled and disabled; auto returns 400.
• In our tests with “Explain idempotency in one sentence”, Seed 2.1 Pro thought for about 1,500 tokens by default and took about 40 seconds; Seed Evolving used about 1,700 tokens and took about 54 seconds. With thinking off, responses came back in about 2 seconds.
• Thinking tokens count toward max_tokens. If the limit is too low, content may be empty, with finish_reason set to length.
• Thinking tokens are billed at the output rate; the count is in usage.completion_tokens_details.reasoning_tokens.

Reading thinking content: in streaming responses it is in delta.reasoning_content; in non-streaming responses it is in message.reasoning_details, where you concatenate the text of each item in order.

Use in Claude Code

Seed 2.1 Pro, Seed 2.1 Turbo and Seed Evolving support the Anthropic-native API at https://api.ofox.ai/anthropic:

• On the Anthropic API, no thinking content is returned unless you pass the thinking parameter. To get thinking, pass "thinking": {"type": "enabled", "budget_tokens": 2000} or {"type": "adaptive"}; thinking content is returned as thinking blocks.
• Seed Character does not support the Anthropic API; calls return protocol_not_supported.

For full Claude Code setup steps, see Claude Code setup.

Image input and structured output

• Image input: all of the models above support image input, using the standard image_url content block; see Vision for the syntax. Image URLs must be quickly downloadable from the public internet, or the request returns Timeout while downloading url; you can also send base64 data instead.
• Structured output: response_format supports json_schema and json_object; see Structured output for the syntax.
• Tool calling: uses the standard tools parameter and returns tool_calls.

Common errors

For other general errors, see Error Handling.

Official docs

• Volcano Engine Ark · Docs 

FAQ