Back to Models

Doubao Seed 2.1 Pro

bytedancebytedance/doubao-seed-2.1-pro

ByteDance Doubao Seed 2.1 Pro β€” frontier model with text + image input. 256K context. Strong coding and agent performance. Cost Β₯6/$0.833 / Β₯30/$4.167 per MTok (I/O).

Doubao Seed 2.1 Pro is ByteDance's June 2026 flagship Seed-family model for coding, long-horizon agents, and multimodal understanding. TheRouter exposes it as bytedance/doubao-seed-2.1-pro: a 256K-context chat model with text and image input, text output, OpenAI-style tools and JSON response formatting, and a 32,768-token maximum completion budget. ByteDance positioned the launch as a production-grade step change rather than another chatbot refresh, with emphasis on code delivery, GUI / image understanding, long-chain task execution, and enterprise stability.

For TheRouter operators, Seed 2.1 Pro is the ByteDance route to test when a workload sits between frontier coding models and high-volume Chinese enterprise agents. It is cheaper than the Western frontier tier while keeping vision input and tool-agent affordances in the same OpenAI-compatible /v1/chat/completions integration. Use it as a primary route for Chinese-first coding and agent traffic, or as a cost-controlled alternative to Claude / GPT-class models when the acceptance test is tool completion, repository edits, document+image reasoning, or multi-step browser / GUI control rather than pure English prose polish.

Best for
  • β€’ Chinese-first coding agents that need repository-scale context, structured output, tool calls, and lower token cost than Western frontier coding routes.
  • β€’ Long-chain enterprise agents β€” planning, browser / GUI control, workflow automation, and tool orchestration where 256K context helps preserve state.
  • β€’ Document-plus-image reasoning, screenshot QA, UI understanding, and multimodal support flows that require text output rather than image generation.
  • β€’ Cost-sensitive fallback behind Claude Opus / Sonnet or GPT-class agents when acceptance tests show comparable tool-task completion on Chinese or mixed-language workloads.
Reach for something else if
  • β€’ Pure image, video, or audio generation β€” this route reads images but returns text; use Seedream, Seedance, GPT Image, or audio models for generative media output.
  • β€’ Workloads that require open weights, self-hosting, or fine-tuning control β€” Doubao Seed 2.1 Pro is a proprietary API model.
  • β€’ Ultra-low-latency commodity chat at massive volume β€” route to Doubao Seed 2.1 Turbo, Seed Evolving, or a smaller Chinese model when Pro-level reasoning is unnecessary.
  • β€’ Regulated data flows that cannot leave a specific residency boundary until your account-level TheRouter and Volcano Engine routing / logging regions are contractually verified.

How TheRouter serves this differently from the vendor

As the vendor operates it

ByteDance operates Doubao Seed 2.1 Pro natively through Volcano Engine Ark, which exposes its own SDK and REST surface (doubao-seed-2-1-pro-260628), with Volcano Engine account management, token billing in CNY, and separate capacity pools for enterprise and consumer Doubao workloads.

On TheRouter

TheRouter re-exposes the model as an OpenAI-compatible /v1/chat/completions endpoint under the public id bytedance/doubao-seed-2.1-pro. Billing is in USD at TheRouter rates, authentication uses a TheRouter API key, and the Volcano Engine Ark upstream is abstracted away entirely. Capabilities are limited to what TheRouter maps: streaming, tools, JSON response_format, vision input, and the supported-parameter set in the catalog; Volcano Ark-native features outside that mapping (e.g. CNY billing controls, Ark-specific system prompts, or enterprise-tier capacity reservation) are not accessible through the TheRouter route.

Context Length
256K
Max Output
33K
Input Priceper 1M tokens
$0.9612/ 1M tokens
Output Priceper 1M tokens
$4.81/ 1M tokens

Modalities

textimage→text

Capabilities

Vision

Pricing Breakdown

TypeRate
Input$0.9612 / 1M tokens
Output$4.81 / 1M tokens

Supported Parameters

temperaturemax_tokenstop_ptoolstool_choiceresponse_formatstop

Specifications

Release date2026-06-23news.qq.com β†—verified
TheRouter model idbytedance/doubao-seed-2.1-proTheRouter model catalog β†—verified
Upstream modeldoubao-seed-2-1-pro-260628 via Volcano Engine Arkprovider-models/volcengine-ark.yaml β†—verified
Context window256,000 tokensstandard-models.yaml β†—verified
Maximum completion32,768 tokensstandard-models.yaml β†—verified
Input / output modalitiesText and image in; text outstandard-models.yaml β†—verified
Supported OpenAI-style parameterstemperature, max_tokens, top_p, tools, tool_choice, response_format, stopstandard-models.yaml β†—verified
TheRouter price$1.104 / 1M input tokens; $5.496 / 1M output tokensTheRouter model catalog β†—verified
Volcano Engine public priceΒ₯6 / 1M input tokens; Β₯30 / 1M output tokens; cache-hit price reported at Β₯1.2 / 1M tokensvolcengine.com/product/doubao β†—verified
Primary capability focusGeneral agent execution, code engineering delivery, multimodal and visual understanding, and video comprehensionseed.bytedance.com β†—verified
License / weightsProprietary API model; weights are not publishedvolcengine.com β†—verified
Training cutoffNot publicly disclosedunknown

Benchmarks

BenchmarkDistributionScoreSource
Terminal Bench 2.1
73.8 score for Seed 2.1 Pro on Terminal Bench 2.1 (Terminal Usage category) per the official ByteDance Seed 2.1 evaluation results page.
73.8seed.bytedance.com β†—
SWE-Atlas
44.7 score for Seed 2.1 Pro on SWE-Atlas (Debugging category) per the official ByteDance Seed 2.1 evaluation results page. Earlier secondary sources referred to this benchmark as 'SWE-Pro'; the official table uses 'SWE-Atlas'.
44.7seed.bytedance.com β†—
SciCode
SciCode was cited in secondary launch coverage as a benchmark where Seed 2.1 Pro performs in the top group, but SciCode does not appear in the official ByteDance Seed 2.1 evaluation results table; a numeric score is not publicly disclosed in any vendor-official source retrieved.
β€”Not publicly disclosedβ€”
MMMU-Pro (w. Tool)
81.2 score for Seed 2.1 Pro on MMMU-Pro with Tool (MultiModal STEM category) per the official ByteDance Seed 2.1 evaluation results page. A parenthetical 82.7 appears alongside, consistent with an extended-compute variant.
81.2 (82.7 with extended compute)seed.bytedance.com β†—
OSWorld / MobileWorld
OSWorld and MobileWorld were cited in secondary launch coverage as agent evaluations where Seed 2.1 Pro leads globally, but neither appears in the official ByteDance Seed 2.1 evaluation results table retrieved; numeric scores are not publicly disclosed in any vendor-official source.
β€”Not publicly disclosedβ€”
Code Arena frontend leaderboard
A score of 1539 and rank 8 were reported by a secondary source (datanorth.ai) for a Seed 2.1 Pro preview. No official ByteDance source publishes Code Arena results for this model; the figure cannot be confirmed from vendor-official material.
β€”Not publicly disclosedβ€”

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": "bytedance/doubao-seed-2.1-pro",
    "messages": [
      {"role": "user", "content": "Summarize the key points from this input."}
    ]
  }'

Chat completion

Drop the model into an OpenAI-compatible chat client for long-context Chinese / mixed-language coding and agent work.

cURL
curl https://api.therouter.ai/v1/chat/completions \
  -H "Authorization: Bearer $THEROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model":"bytedance/doubao-seed-2.1-pro","messages":[{"role":"system","content":"You are a precise TypeScript code reviewer."},{"role":"user","content":"Review this diff for correctness and return JSON with risks and fixes."}],"response_format":{"type":"json_object"}}'

More from bytedance

Similar models

Cross-provider sibling models
doubao/doubao-seed-2-0-pro
Previous Doubao flagship exposed through the Doubao provider namespace.
qwen/qwen3-coder-plus

Not in the current catalog; the description below is historical.

Alternative Chinese coding-specialist route for repo and agent workflows.

Browse the current model catalog
anthropic/claude-sonnet-4.6
Western frontier baseline for coding and agent eval comparisons.

News & changes

2026-06-23

ByteDance launches Doubao 2.1 with Pro, Turbo, and Evolving routes

Volcano Engine presented Doubao Seed 2.1 Pro as the flagship model for coding, agent, and VLM workloads, with API availability through Volcano Engine and a Turbo sibling priced at half the Pro tier.

re-authored by TheRouternews.qq.com β†—
2026-06-24

Seed 2.1 Pro positioned against Claude Opus 4.6 on coding and agent cost

A secondary launch summary reports ByteDance's claim that Seed 2.1 Pro targets coding, long-chain agent work, and multimodal reasoning while reducing total cost of ownership by roughly 80% versus Claude Opus 4.6. Treat that as vendor positioning until workload-specific evals confirm it.

re-authored by TheRouterdatanorth.ai β†—

Frequently asked

Is Doubao Seed 2.1 Pro OpenAI-compatible on TheRouter?

Yes. Call https://api.therouter.ai/v1/chat/completions with the standard OpenAI SDK and set model to bytedance/doubao-seed-2.1-pro. TheRouter catalog lists support for temperature, max_tokens, top_p, tools, tool_choice, response_format, and stop.

re-authored by TheRouterstandard-models.yaml β†—
Does Doubao Seed 2.1 Pro support image input?

Yes. TheRouter's catalog exposes text and image input with text output. Use it for screenshot, document image, chart, and UI understanding tasks; use a dedicated image-generation route if you need pixels back.

re-authored by TheRouterTheRouter model catalog β†—
How should I choose between Doubao Seed 2.1 Pro and Turbo?

Start with Pro for difficult coding, agent, and multimodal reasoning tasks where success rate matters. Move traffic to Turbo when prompts are simpler, text-only, high-frequency, and your eval harness shows the lower-cost route still passes acceptance tests.

re-authored by TheRouternews.qq.com β†—
Is the published benchmark data enough to promote it straight to production?

No. The public sources retrieved for this page mostly provide vendor-reported leaderboard positioning, not full numeric tables or TheRouter latency data. Promote it only after a workload-specific eval checks tool-call correctness, multimodal error rate, latency, and cost per successful task.

re-authored by TheRouterdatanorth.ai β†—
Fact ledger β€” every claim on this page traces here
sourceURLretrieved
Release datenews.qq.com β†—2026-07-22verified
TheRouter model idTheRouter model catalog β†—2026-07-22verified
Upstream modelprovider-models/volcengine-ark.yaml β†—2026-07-22verified
Context windowstandard-models.yaml β†—2026-07-22verified
Maximum completionstandard-models.yaml β†—2026-07-22verified
Input / output modalitiesstandard-models.yaml β†—2026-07-22verified
Supported OpenAI-style parametersstandard-models.yaml β†—2026-07-22verified
TheRouter priceTheRouter model catalog β†—2026-07-22verified
Volcano Engine public pricevolcengine.com/product/doubao β†—2026-07-22verified
Primary capability focusseed.bytedance.com β†—2026-08-19verified
License / weightsvolcengine.com β†—2026-07-22verified
Training cutoffβ€”β€”unknown
Terminal Bench 2.1seed.bytedance.com β†—2026-08-19verified
SWE-Atlasseed.bytedance.com β†—2026-08-19verified
SciCodenews.qq.com (secondary) β†—2026-07-22unknown
MMMU-Pro (w. Tool)seed.bytedance.com β†—2026-08-19verified
OSWorld / MobileWorldnews.qq.com (secondary) β†—2026-07-22unknown
Code Arena frontend leaderboarddatanorth.ai (secondary) β†—2026-07-22unknown
ByteDance launches Doubao 2.1 with Pro, Turbo, and Evolving routesnews.qq.com β†—2026-07-22verified
Seed 2.1 Pro positioned against Claude Opus 4.6 on coding and agent costdatanorth.ai β†—2026-07-22verified
Is Doubao Seed 2.1 Pro OpenAI-compatible on TheRouter?standard-models.yaml β†—2026-07-22to verify
Does Doubao Seed 2.1 Pro support image input?TheRouter model catalog β†—2026-07-22to verify
How should I choose between Doubao Seed 2.1 Pro and Turbo?news.qq.com β†—2026-07-22to verify
Is the published benchmark data enough to promote it straight to production?datanorth.ai β†—2026-07-22to verify
Help & contact