Overview
Doubao 2.1 Pro (Doubao-Seed 2.1 Pro) is a new-generation flagship model officially released by ByteDance's Volcano Engine at the **FORCE Original Power Conference** on 2026-06-23. It marks ByteDance's transition from a "model provider" to an "AI ecosystem platform." **Pro coding capabilities are on par with Claude Opus 4.7**, and overall performance rivals GPT-5.5 / Gemini 3.1 Pro—this is the first time a domestic model has entered the international top-tier closed-source flagship range in long-range coding and general benchmarks.
Core capabilities boil down to two things: **autonomously advancing a complete engineering project for 10+ consecutive hours** (at the level of chip RTL design and full-stack software projects) + **watching a 2-hour video and performing end-to-end editing**—the latter is the first domestic model to string together long video understanding and editing into a complete workflow.
**Pricing is the highlight of this release**: Pro input **¥6 / output ¥30 per million tokens** (cache hit ¥1.2), **Turbo is half the price of Pro**, **overall usage cost is nearly 80% lower than the Claude Opus 4.6-4.8 series**—ByteDance officially states "computational cost for the same task is nearly 80% lower than overseas closed-source models." Combined with Doubao's **daily token call volume of 180 trillion** (a 1500x increase in two years) and **public cloud market share rising to 49.5%**, the dual fronts of price and scale squeeze overseas closed-source models.
This is a landmark event in 2026—"**ByteDance flips the table**"—36Kr's headline that day directly read "ByteDance flips the table, Doubao 2.1 costs slashed by 80%."
Key Features
- Coding on par with Opus 4.7: Pro coding capabilities are on par with Claude Opus 4.7, the first time a domestic model enters the international top-tier closed-source flagship range
- 10+ hour autonomous engineering: Can autonomously advance a complete engineering project for 10+ consecutive hours (at the level of chip RTL design and full-stack software projects)
- Watch 2-hour video + end-to-end editing: Can watch a 2-hour video and directly perform end-to-end editing, the first domestic model to string together long video understanding and editing into a workflow
- Cost reduced by nearly 80%: Overall usage cost is nearly 80% lower than the Opus 4.6-4.8 series, benefiting from ByteDance's self-developed models and proprietary computing power
- Three tiers (Pro/Turbo/Seed-Evolving): Pro flagship, Turbo cost-effective, Seed-Evolving continuous learning version, covering all scenarios
- Multimodal depth: Significantly improved accuracy in image-text, screenshot, and table parsing; interleaved reasoning across images/video/text
- 180 trillion daily token scale: Doubao's daily token call volume is 180 trillion, a 1500x increase in two years, with 49.5% public cloud market share
- FORCE ecosystem family bucket: Released simultaneously with Seedance 2.5 (video) + Seedream 5.0 (image) + Doubao Audio 1.0, forming a multimodal synergy matrix
Use Cases
- 10+ hour long-range coding tasks (replacing Claude Code + Opus 4.7 backend)
- Chip RTL design, full-stack software project-level autonomous engineering
- Long video understanding + end-to-end editing (film pre-production, educational courses, training videos)
- Cost-sensitive SME AI platform selection
- Domestic model replacement in state-owned enterprises / Xinchuang scenarios
- Default upgrade for deep users of ByteDance Volcano Engine / Feishu ecosystem
Pros
- Coding on par with Opus 4.7, overall performance rivals GPT-5.5 / Gemini 3.1 Pro
- Cost reduced by nearly 80%, the most impactful move in the 2026 domestic large model price war
- The combination of 10+ hour autonomous engineering + 2-hour video editing is unique domestically
- 180 trillion daily tokens validates product scalability
- Three tiers (Pro/Turbo/Seed-Evolving) cover all scenarios
- FORCE ecosystem family bucket (video + image + audio + coding) synergy
Pricing
**Pro**: Input ¥6/M token, Output ¥30/M token, **Cache hit ¥1.2/M**.\n**Turbo**: Half the price of Pro.\n**Overall cost** nearly 80% lower than the Opus 4.6-4.8 series.\n**Accessed via Volcano Engine / Doubao API / Feishu AI**.
Summary
Doubao 2.1 Pro is a landmark event in 2026—"ByteDance flips the table"—**coding on par with Opus 4.7 + 10+ hour autonomous engineering + watch 2-hour video editing + cost reduced by nearly 80%** gives it real competitiveness against overseas closed-source flagships. If you work on long-range coding or long video content, Doubao 2.1 Pro is one of the most worth-trying domestic options; if you are a deep user of ByteDance Volcano Engine / Feishu ecosystem, it's almost a no-brainer upgrade; if you do serious Chinese NLP / knowledge work, it can complement DeepSeek V4 / Kimi K2.7 Code / GLM-5.2; if you focus on overseas or English-language tasks, GPT-5.6 / Claude series remain the primary choice.
Version History
- 官网最新动态显示豆包大模型1.6升级（国内首个原生支持分档调节思考长度）、豆包语音模型重磅升级、方舟模型服务平台发布智能 (2025-10-16): The latest updates on the official website show that Doubao Large Model 1.6 has been upgraded (the first in China to natively support tiered adjustment of thinking length), Doubao voice model has undergone a major upgrade, and the Ark model service platform has released intelligent model routing features. These updates all came after the date of the 2.1 Pro version mentioned in the evaluation, and are new developments not covered by the evaluation.
- Doubao 2.1 Pro / Turbo / Seed-Evolving (2026-06-23): Coding on par with Opus 4.7, 10+ hour autonomous engineering, watch 2-hour video editing; cost reduced by 80%; simultaneously released Seedance 2.5 + Seedream 5.0 + Doubao Audio 1.0
- Doubao 2.0 (2026-02-14): Comprehensive upgrade of multimodal AI capabilities, reshaping a new paradigm for industrial intelligence
- Doubao 1.x (2024-2025): Evolution from a single-point assistant to a multimodal platform