Skip to content
#02S+Feb 15, 2026Evidence A

Doubao Seed 2.0 Pro

ByteDance · Undisclosed (MoE)

The practical ceiling for Chinese multimodal work

Topped the April 2026 SuperCLUE-VLM board at 90.66, with 96.30 basic cognition and 92.82 visual reasoning. Video timing and Chinese context are its moat; price is the tax.

TextImageVideoAudioNative

Last updated Oct 18, 2026Editorial scoring

Bottom line

If your workload is Chinese content, short video and real phone scenes, and budget allows, Seed 2.0 Pro is the safest default. It does not win every single dimension, but it has no glaring hole and holds long tasks better than most rivals.

Eight-dimension breakdown

editorial mapping · 0–100
Perception97Document96Chart QA94Video96Agent93Generation88Chinese97Ecosystem988D
  • Doubao Seed 2.0 Pro95.2
Visual perception & grounding20%
97
Document parsing & Chinese OCR18%
96
Chart & document reasoning14%
94
Video & long-horizon reasoning14%
96
GUI agents & tool use14%
93
Generation & editing6%
88
Chinese context & culture6%
97
Ecosystem, cost & availability8%
98

Weights: perception 20 / documents 18 / charts 14 / video 14 / agents 14 / generation 6 / Chinese 6 / ecosystem 8. Definitions and sources on the methodology page.

Public benchmarks

as reported, not converted
Benchmark / metricValueNote
SuperCLUE-VLM 2026-0490.66#1 of 17 models
基础认知96.30SuperCLUE-VLM sub-score
视觉推理92.82SuperCLUE-VLM sub-score

Public scores come from different versions and prompts and are not interchangeable with our composite; they are one evidence input.

Scenario mapping

five stress tests, one scale
ScenarioScoreObservation
Chinese receipt OCR95 · Occasional misses on handwriting and creased receipts
Chart data extraction93 · Stable on stacked multi-series charts
30-min video QA96 · Strong event localization and AV alignment
Mobile GUI control92 · Accurate taps; occasional backtracking on long tasks
Chinese cultural context97 · Excellent on heritage, solar terms and memes

Strengths and limitations

Strengths

  • Strongest Chinese priors: dishes, memes, heritage sites and policy text are notably more stable zero-shot.
  • Leads on temporal video understanding; event localization drifts less on long clips.
  • Mature API, SDK, Volcano Engine ecosystem and rate limits; low integration cost.

Limitations

  • At ¥15/1M output tokens, high-frequency image analysis gets expensive.
  • Hard math and heavy coding still trail DeepSeek and Kimi text flagships.
  • Closed weights, no self-hosting; regulated data requires a dedicated deployment deal.

Price and real cost

List price

¥15/M tokens

≈ ¥15 / 1M output tokens. For 1,000 images at 1,024px, ~1,100 image tokens each and 300 output tokens, the monthly cost is about ¥9.1 (assuming input at 25% of output price).

Use your volume

About Doubao Seed 2.0 Pro

Is Doubao Seed 2.0 Pro open source?

Licence: Closed commercial API. No public weights; access is via vendor API or an on-prem licence.

How much does Doubao Seed 2.0 Pro cost?

≈ ¥15 / 1M output tokens. It ranks #2 with a composite of 95.2. Real bills depend on resolution, tiling, caching and volume discounts.

What is Doubao Seed 2.0 Pro best for?

Chinese short-video moderation and tagging; Mobile GUI agents and automation; Chinese e-commerce image understanding。

Why does Doubao Seed 2.0 Pro score 95.2?

The composite weights eight dimensions: perception 20%, documents 18%, charts 14%, video 14%, agents 14%, generation 6%, Chinese 6%, ecosystem 8%. Its strongest dimension is Ecosystem, cost & availability; its weakest is Generation & editing.

Personal ranking · Not an official score

Every ranking, score and comparison on this page is an independent personal ranking, not an official score, and represents no vendor, institution or regulator. Scores are a subjective mapping of public evidence under our own weights — change the weights and the order changes. Disagree, or spotted a factual error? Contact the site owner: jizhi2024@foxmail.com — we check each report and log corrections in the changelog.