Doubao Seed 1.6

doubaodoubao/doubao-seed-1-6

Doubao Seed 1.6 — ByteDance Seed-series next-gen general model with tools and structured output.

Volcano Engine Ark model docs ↗ByteDance Seed 1.6 announcement ↗

Doubao-Seed-1.6 is ByteDance's flagship general-purpose model released June 11, 2025 at the Volcano Engine FORCE Conference. It introduces adaptive thinking modes (auto / thinking / non-thinking) and significantly improved reasoning, instruction following, and agent capabilities over the 1.5 generation. The model supports tools, structured outputs, and multimodal understanding in its vision and flash siblings.

For TheRouter operators, Doubao-Seed-1.6 is the primary Chinese-first reasoning endpoint that offers competitive math and coding performance at dramatically lower cost than frontier Western models. The thinking mode can be toggled via the non-standard thinking parameter ({type: "auto"|"thinking"|"non-thinking"}) — LiteLLM issue 11879 documents this extension. The model is reachable through api.therouter.ai/v1 with full OpenAI chat/completions compatibility plus the ByteDance-specific thinking extension.

Best for

• Chinese-first reasoning, math, and coding workloads where cost-efficiency and native language quality matter more than absolute English benchmark leadership.
• Agent loops that need reliable tool calling + structured output at sub-frontier pricing with optional deep-thinking mode for hard sub-problems.

Reach for something else if

• Pure English high-stakes reasoning or creative writing where Claude Sonnet 4.6 / GPT-4o currently lead measured English benchmarks and human preference studies.
• Native video generation or audio synthesis — use doubao-seedance-2-0 / doubao-seedream-5-0 siblings instead.

Context Length

131K

Max Output

33K

Input Priceper 1M tokens

$0.270/ 1M tokens

Output Priceper 1M tokens

$2.16/ 1M tokens

Modalities

text→text

Pricing Breakdown

Type	Rate
Input	$0.270 / 1M tokens
Output	$2.16 / 1M tokens

Supported Parameters

temperaturemax_tokenstop_ptoolstool_choiceresponse_formatstop

Specifications

Release date	2025-06-11ainativefoundation.org ↗	verified
Architecture	Next-generation Seed-series MoE with adaptive thinking modes (auto/thinking/non-thinking)www.volcengine.com ↗	verified
Input modalities	Text (vision variant adds image+video)www.aisharenet.com ↗	verified
License	Closed (API only)	verified
Training cutoff	Not publicly disclosed	unknown
Thinking modes	auto, thinking, non-thinking (via thinking.type param)github.com ↗	verified

Benchmarks

Benchmark	Distribution	Score	Source
GPQA Diamond Thinking variant; reported at 1.6 launch		81.5%	ainativefoundation.org ↗
AIME 2025 Thinking variant; reported at 1.6 launch		86.3%	ainativefoundation.org ↗

API Usage Examples

Use the global api.therouter.ai endpoint shown below for new integrations; the legacy China accelerated endpoint is retired.

cURL

curl https://api.therouter.ai/v1/chat/completions   -H "Content-Type: application/json"   -H "Authorization: Bearer $THE_ROUTER_API_KEY"   -d '{
    "model": "doubao/doubao-seed-1-6",
    "messages": [
      {"role": "user", "content": "Summarize the key points from this input."}
    ]
  }'

API guide

Full API reference →

Chat completion

cURL

curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"doubao/doubao-seed-1-6","messages":[{"role":"user","content":"Solve: 2x+3=11"}],"tools":[{"type":"function","function":{"name":"calc","description":"...","parameters":{}}}]}'

More from doubao

doubao/doubao-seed-1-6-flashUltra-low-latency variant doubao/doubao-seed-1-6-visionVision-language extension doubao/doubao-seed-1-8Next incremental upgrade

Similar models

Cross-provider sibling models

deepseek/deepseek-v3.2

Open-weight Chinese reasoning alternative

qwen/qwen3-235b

Alibaba Qwen3 flagship reasoning model

anthropic/claude-sonnet-4

Western frontier reasoning reference

News & changes

2025-06-11

ByteDance launches Doubao Seed 1.6 with adaptive thinking

Doubao-Seed-1.6 family (including thinking and flash variants) introduced at Volcano Engine FORCE Conference. Thinking variant scores 81.5 GPQA Diamond / 86.3 AIME25; 256K context window marketed by third-party hosts.

re-authored by TheRouterainativefoundation.org ↗

Frequently asked

How do I enable deep thinking on Doubao-Seed-1.6?

Pass the non-standard thinking: {type: "thinking"} (or "auto") object in the request body. The parameter is documented by LiteLLM and accepted by the Volcano Engine backend; TheRouter forwards it transparently.

re-authored by TheRoutergithub.com ↗

What is the difference between doubao-seed-1-6, doubao-seed-1-6-thinking and doubao-seed-1-6-flash?

doubao-seed-1-6 is the balanced general model. doubao-seed-1-6-thinking forces deeper chain-of-thought for hard reasoning/math. doubao-seed-1-6-flash is the ultra-low-latency variant optimized for real-time chat and high-throughput agent loops.

re-authored by TheRouterwww.aisharenet.com ↗

Fact ledger — every claim on this page traces here

source	URL	retrieved
Release date	ainativefoundation.org ↗	2026-05-26	verified
Architecture	www.volcengine.com ↗	2026-05-26	verified
Input modalities	www.aisharenet.com ↗	2026-05-26	verified
License	—	—	verified
Training cutoff	—	—	unknown
Thinking modes	github.com ↗	2026-05-26	verified
GPQA Diamond	ainativefoundation.org ↗	2026-05-26	to verify
AIME 2025	ainativefoundation.org ↗	2026-05-26	to verify
ByteDance launches Doubao Seed 1.6 with adaptive thinking	ainativefoundation.org ↗	2026-05-26	verified
How do I enable deep thinking on Doubao-Seed-1.6?	github.com ↗	2026-05-26	to verify
What is the difference between doubao-seed-1-6, doubao-seed-1-6-thinking and doubao-seed-1-6-flash?	www.aisharenet.com ↗	2026-05-26	to verify