NezhaGateNezhaGate
简体中文English ✓Español한국어日本語Русский
Log in Get API Key
Chat

Doubao Seed 2.1 Pro

⧉
doubao-seed-2-1-pro
Model Type ProTurbo

Doubao Seed 2.1 Pro is the flagship of ByteDance's Doubao Seed 2.1 series, built for reasoning, coding and agent work, with image input. By default it thinks before it answers, with the reasoning in reasoning_content; reasoning_effort: "minimal" usually skips the thinking for a direct answer (8 of 10 calls in our tests, not guaranteed). Tool calling, JSON output and streaming are supported, and images can be sent as base64 or public links. Fully OpenAI-compatible: set model to doubao-seed-2-1-pro. Pay-as-you-go, failed calls never billed.

ChatReasoningCodingImage inputOpenAI compatible
Get API Key → API Docs

Live Test · Playground

Try out Doubao Seed 2.1 Pro right here (available after login).

Input

Advanced
Web search
Memory · multi-turn

Conversation

Start a conversationType a message below to begin

About Doubao Seed 2.1 Pro

Doubao Seed 2.1 Pro is the flagship of ByteDance's Doubao Seed 2.1 series, served on NezhaGate through the OpenAI-compatible API. It is built for reasoning, coding and agent work, and it accepts image input. By default it thinks before it answers: the reasoning comes back in message.reasoning_content and the answer in content; reasoning_effort: "minimal" usually skips the thinking for a direct answer, though not every time. Tool calling, JSON output and streaming are supported. Set model to doubao-seed-2-1-pro; pay-as-you-go, failed calls never billed.

Use cases

Complex reasoning and maths

It thinks before it answers by default, which suits multi-step reasoning, maths and logic problems; the reasoning is visible in reasoning_content.

Code and agents

OpenAI-format tools / tool_calls and JSON output plug into any agent framework for retrieval, function calls and multi-step tasks.

Image understanding

Q&A over screenshots, receipts and charts; in our tests it read the text in an image accurately. Send images as base64 or public links.

Fast, answer-only replies

reasoning_effort: "minimal" usually skips the thinking (8 of 10 calls in our tests), which suits simple tasks where you only need the answer, with fewer output tokens.

How to choose

Choose Pro for the flagship reasoning, coding and agent work of the Doubao Seed 2.1 series; choose Turbo for everyday chat, high-volume jobs or when the unit price matters most: it costs about half as much per token and caches a repeated long prefix automatically. Both belong to the Doubao Seed 2.1 family, so switching is just the model field.

FAQ

Can I turn thinking off?
Usually, but not guaranteed. With reasoning_effort: "minimal" the model mostly answers directly: in our tests 8 of 10 calls skipped the thinking, while the other 2 still returned reasoning. thinking: {"type": "disabled"} has no effect on this model. When omitted, the model thinks first even on trivial prompts (a five-word greeting used 362-802 tokens in our tests), and reasoning tokens bill at the output rate inside usage.completion_tokens.
Can max_tokens limit the output?
No. max_tokens is accepted but does not cut the output off (with max_tokens 100, a five-word greeting still produced 362-802 tokens in our tests), and you are billed for the actual tokens in usage. For short answers, ask for brevity in the prompt; reasoning_effort: "minimal" usually helps too.
Is prompt caching supported?
With cache_control {"type": "ephemeral"} on a content part, a repeated long prefix can hit the cache: in our tests all 9,827 input tokens of the second call were cached, and the cached part settles at the cache rate. Without the marker we saw no hits. Whether it hits is decided upstream, and billing follows the returned usage. If you want automatic caching, choose Turbo.
What should I know about long answers?
The model thinks first, and a one-paragraph answer took 50-90 seconds in our tests. For long output or latency-sensitive calls use stream=true: the reasoning streams in delta.reasoning_content and the answer in delta.content, while a non-streaming request that waits too long may be cut off.
Am I charged for failed calls?
No. Only calls that return normally are settled, at actual usage; upstream errors and timeouts never touch your balance.

Related Models

Explore other models you can integrate.

View all →
Doubao Seed 2.1 Turbo Chat

Doubao Seed 2.1 Turbo

doubao-seed-2-1-turbo

The lower-cost Doubao Seed 2.1 - everyday chat and high volume

$0.36/1M in · $1.79/1M out ↓20% View →
Qwen3.8 Max Chat

Qwen3.8 Max

qwen3.8-max

Alibaba's Qwen 3.8 flagship - reasoning, coding and agents

$1.40/1M in · $4.20/1M out ↓30% View →
Kimi K3 Chat

Kimi K3

kimi-k3

Moonshot's new flagship - long context and agents

$2.10/1M in · $10.50/1M out ↓30% View →
GLM-5.3 Chat

GLM-5.3

glm-5.3

The new Z.ai flagship - coding and agents

$0.34/1M in · $1.25/1M out ↓73% View →
DeepSeek V4.1 Flash Chat

DeepSeek V4.1 Flash

deepseek-v4.1-flash

DeepSeek's current Flash model - thinking on or off

$0.09/1M in · $0.36/1M out ↓40% View →