How to Use the Seed 2.1 Turbo API

Seed 2.1 Turbo is available on EmpirioLabs. It is ByteDance's next-generation model for the coding and agent era, with engineering-grade code delivery, long-horizon agentic execution, and multimodal understanding across text, image, and video. It runs on an OpenAI-compatible API with a 256K context window.

What Seed 2.1 Turbo is good at

Seed 2.1 Turbo is tuned for real software work: understanding a repository, making changes, testing them, and repairing its own output across long, multi-step tasks. It reasons before answering, calls tools, returns structured JSON on demand, and reads images and video, so it fits coding assistants, autonomous agents, and multimodal analysis in a single endpoint.

How to call it

Send a standard chat completion. The model reasons by default, so a considered answer needs no extra setup:

curl https://api.empiriolabs.ai/v1/chat/completions \
  -H "Authorization: Bearer $EMPIRIOLABS_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "seed-2-1-turbo",
    "messages": [
      {"role": "user", "content": "Refactor this function to remove the nested loop, then explain the change."}
    ],
    "reasoning_effort": "high"
  }'

It also works on /v1/responses, /v1/messages, and the Gemini-compatible generateContent route, so you can reach it from the OpenAI, Anthropic, and Google SDKs by pointing the base URL at EmpirioLabs.

Reasoning, tools, and structured output

Reasoning is on by default at high effort. Lower it with reasoning_effort (low, medium, high) for faster replies, or set enable_thinking to false to answer directly. The model supports standard function calling with parallel tool calls, and JSON output through response_format (JSON mode or a strict JSON schema). Turn on enable_web_search to let it pull live web sources before answering.

Image and video input

Seed 2.1 Turbo reads images and video alongside text. Pass a public image or video URL directly, no base64 required, and use image_detail and video_fps to control how closely it looks:

{
  "model": "seed-2-1-turbo",
  "messages": [
    {"role": "user", "content": [
      {"type": "text", "text": "What does this screen show, and what would you click next?"},
      {"type": "image_url", "image_url": {"url": "https://example.com/screenshot.png"}}
    ]}
  ]
}

Tarification

Seed 2.1 Turbo is pay-as-you-go per token, with no subscription. Unusually for the Seed family, its per-token rate is flat across the full 256K context, with no higher rate for long inputs. See the live rate on the model page and the pricing page.

Bon à savoir

• Reasoning defaults to high effort, so responses include a thinking pass. Lower reasoning_effort or disable thinking when you want speed over depth.
• Pricing is flat across the whole 256K context, so long-context requests do not jump to a higher rate.
• Image and video inputs are fetched from public URLs, so you can pass links directly instead of encoding files.

Try it in the playground or read the API docs.