> ## Documentation Index
> Fetch the complete documentation index at: https://docs.apiyi.com/llms.txt
> Use this file to discover all available pages before exploring further.
# Seed 2.1 Turbo Responses API 레퍼런스
> Seed 2.1 Turbo (dola-seed-2-1-turbo-260628) Responses API 레퍼런스 및 대화형 플레이그라운드: 네이티브 멀티턴 previous_response_id와 연결된 명시적 캐싱입니다.
오른쪽 플레이그라운드를 사용하십시오: **인증**에 `Bearer sk-your-api-key`을 넣고, `input`를 입력한 다음 전송하십시오. 응답의 `output` 배열에서, `type: "reasoning"` 항목은 추론 요약이며 — `type: "message"`는 실제 답변입니다.
멀티턴 대화에서는 이전 응답의 `id`(`resp_` 접두사)를 다음 호출의 `previous_response_id`로 전달하면 되며 — 히스토리를 다시 보낼 필요가 없습니다. 명시적 캐싱과 추론 제어는 [Seed 2.1 Turbo 개요](/ko/api-capabilities/dola-seed-2-1-turbo/overview)에서 다룹니다.
* 작은 `max_output_tokens`는 추론에 잠식되어 **빈 텍스트**의 `status: "incomplete"`를 반환합니다 — 1500부터 시작하십시오
* 명시적 캐싱 `caching: {"type": "enabled"}`은 `previous_response_id`를 통해 **연결된** 경우에만 캐시 적중합니다; 동일한 접두사를 단순히 반복하는 것만으로는 절대 적중하지 않으며(암시적 접두사 캐시도 비활성화됩니다)
## 매개변수 빠른 참조
| 매개변수 | 유형 | 필수 | 기본값 | 비고 |
| ---------------------- | -------------- | -- | ---------- | ----------------------------------------------------------------- |
| `model` | string | ✓ | — | 고정: `dola-seed-2-1-turbo-260628` |
| `input` | string / array | ✓ | — | 문자열 또는 메시지 배열(`function_call_output` 항목 포함) |
| `max_output_tokens` | int | | — | thinking을 포함한 출력 예산; 1500 이상 권장 |
| `reasoning.effort` | string | | — | `low` / `medium` / `high`; 약 \~371(낮음) 대비 \~1317(높음) 추론 token 측정값 |
| `previous_response_id` | string | | — | 기본 멀티턴을 위한 이전 응답 ID입니다. 전체 컨텍스트의 명시적 캐시 적중을 활성화합니다 |
| `caching.type` | string | | `disabled` | `enabled`가 명시적 캐싱을 켭니다(체이닝 필요) |
| `store` | bool | | `true` | 나중에 참조할 수 있도록 이 응답을 저장합니다 |
| `stream` | bool | | `false` | SSE 이벤트 스트림(`response.created` → `response.completed`) |
| `text.format` | object | | — | 구조화된 출력, `json_schema` + `strict` 지원 |
| `tools` | array | | — | 함수 호출 도구 목록(평면 형식) |
## 응답 하이라이트
* `status`가 `incomplete`일 때는 `incomplete_details.reason`을 확인합니다(보통 `length`입니다: 추론이 예산을 소진했습니다)
* 추론 사용량: `usage.output_tokens_details.reasoning_tokens`
* 캐시 적중: `usage.input_tokens_details.cached_tokens`(연속된 2번째 턴은 테스트에서 이전 컨텍스트 전체를 적중했고, 지연 시간이 대략 절반으로 줄었습니다)
* 응답의 `caching` 필드는 실제로 적용된 캐시 모드를 반영합니다
## OpenAPI
````yaml api-reference/dola-seed-2-1-turbo-responses-openapi-en.yaml POST /v1/responses
openapi: 3.1.0
info:
title: Seed 2.1 Turbo Responses API
description: >
ByteDance Seed 2.1 Turbo (`dola-seed-2-1-turbo-260628`) — native Responses
API endpoint.
- Full Responses semantics: event stream, reasoning items, multi-turn via
`previous_response_id`
- **Explicit caching**: `caching: {"type": "enabled"}` combined with
`previous_response_id` chaining hits the entire previous context on turn 2
(measured: 7873 cached tokens, ~2x faster)
- Reasoning is on by default: a small `max_output_tokens` gets eaten by
thinking, yielding `status: incomplete` with empty text - use 1500+
- Supports `reasoning.effort` tiers, structured output (`text.format`), and
function calling
**Authentication**: add `Authorization: Bearer YOUR_API_KEY` header
**Get an API Key**: create a token in the APIYI console at
`api.apiyi.com/token`
version: 1.0.0
servers:
- url: https://api.apiyi.com
description: Primary endpoint
- url: https://vip.apiyi.com
description: Backup endpoint
security:
- bearerAuth: []
paths:
/v1/responses:
post:
tags:
- Text Generation
summary: Responses API with Seed 2.1 Turbo (native multi-turn / explicit cache)
description: >
Call the Responses API with `dola-seed-2-1-turbo-260628`.
- For multi-turn conversations pass `previous_response_id` instead of
resending history
- With `caching.enabled`, cache hits require `previous_response_id`
chaining; merely repeating the same prefix does not hit
operationId: createSeedResponse
requestBody:
required: true
content:
application/json:
schema:
$ref: '#/components/schemas/SeedResponsesRequest'
example:
model: dola-seed-2-1-turbo-260628
input: Introduce yourself in one sentence
max_output_tokens: 1500
responses:
'200':
description: >-
Generation succeeded. The output array contains reasoning and
message items
content:
application/json:
schema:
$ref: '#/components/schemas/SeedResponsesResponse'
'400':
description: Invalid parameters
'401':
description: Unauthorized - invalid API Key
'429':
description: Rate limit exceeded or insufficient balance
'503':
description: >-
Wrong model name or no available channel for your group (this model
returns 503 instead of 404 for unknown model names)
security:
- bearerAuth: []
components:
schemas:
SeedResponsesRequest:
type: object
required:
- model
- input
properties:
model:
type: string
description: Model ID, fixed to dola-seed-2-1-turbo-260628
enum:
- dola-seed-2-1-turbo-260628
default: dola-seed-2-1-turbo-260628
input:
type: string
description: >-
Input content. Either a string or a message array ([{role, content},
...], including function_call_output items)
example: Introduce yourself in one sentence
max_output_tokens:
type: integer
description: >-
Max output tokens (thinking included). Too small yields incomplete
with empty text - use 1500+
default: 1500
reasoning:
type: object
description: >-
Thinking depth control. Measured: effort low ~371, high ~1317
reasoning tokens
properties:
effort:
type: string
enum:
- low
- medium
- high
previous_response_id:
type: string
description: >-
Previous response id (resp_ prefix) for native multi-turn; combine
with caching.enabled for explicit cache hits
caching:
type: object
description: >-
Explicit cache switch. enabled only hits when chained via
previous_response_id
properties:
type:
type: string
enum:
- enabled
- disabled
store:
type: boolean
description: >-
Whether to store this response for later previous_response_id
reference
stream:
type: boolean
description: >-
Stream via SSE (response.created → response.output_text.delta →
response.completed)
default: false
text:
type: object
description: >-
Structured output. Supports {"format": {"type": "json_schema", name,
strict, schema}}
tools:
type: array
description: >-
Function-calling tool list (flat Responses format: {type, name,
description, parameters})
items:
type: object
SeedResponsesResponse:
type: object
properties:
id:
type: string
description: >-
Response ID (resp_ prefix), usable as the next turn's
previous_response_id
status:
type: string
description: >-
completed / incomplete (incomplete when thinking eats the token
budget)
output:
type: array
description: >-
Output items: reasoning (thinking summary), message (text),
function_call, etc.
items:
type: object
usage:
type: object
description: >-
Usage. output_tokens_details.reasoning_tokens = thinking spend;
input_tokens_details.cached_tokens = cache hits
caching:
type: object
description: The cache mode actually in effect for this request
securitySchemes:
bearerAuth:
type: http
scheme: bearer
description: API Key from the APIYI console
````