Baidu AI Cloud Media and Entertainment China Translated from Chinese

NetEase Fuxi

NetEase Fuxi needed a way to move beyond rigid scripted NPC dialogue and support richer real-time interaction for a massive player base without sacrificing world consistency or service stability.

Justice has more than 100 million active players.

Reported impact

Business outcome

NetEase Fuxi delivered a more dynamic game world, expanded what players can ask and do with NPCs, and proved that large-model characters can operate at massive consumer scale.

  • Customer Experience
  • Scalability
  • Speed to Market

Solution

NetEase Fuxi and Baidu AI Cloud brought ERNIE large-model capabilities into the mobile game Justice to create an in-game AI arena populated by interactive NPCs. The solution uses Qianfan AppBuilder and Qianfan ModelBuilder to provide the application and model layer behind AI characters that can answer free-form player questions, stay aligned with their gender, profession, age, geography, and life history, and respond in ways that fit the open-world setting. ERNIE-Character-Fiction-8K is optimized for space-time and scenario consistency so characters can push storylines forward instead of repeating canned scripts. The same platform supports dynamic scene dialogue, broad topic coverage, and real-time interaction at massive scale, while Baidu's infrastructure provides enough concurrency headroom to handle the QPS demands of a title with more than 100 million active players.

Data flow

Player prompts and world context enter the Qianfan-backed NPC system, where the character model interprets context, generates in-character responses, and feeds the interaction back into the live game session. Infrastructure capacity keeps those flows stable even under very high concurrent traffic.

Solution architecture

3 components · 2 layers
  1. Compute
    • Qianfan ModelBuilder Provides the model-development layer used to adapt ERNIE capabilities for game NPC behavior.
    • ERNIE-Character-Fiction-8K Drives real-time character dialogue, scenario consistency, and dynamic story progression for AI NPCs.
  2. Orchestration
    • Qianfan AppBuilder Organizes the in-game AI interaction flows and application logic for NPC conversations.

Architecture clues

  • Baidu provided resource guarantees for high-QPS game traffic
  • NPCs can handle free-form questions rather than only preset dialogue branches
  • The deployment uses a character-specialized ERNIE model optimized for scenario and world consistency

Evidence from the source

Justice has more than 100 million active players and Baidu provided the resource guarantees needed to handle possible QPS challenges smoothly (original: 目前,逆水寒端手游活跃玩家总数已超过1亿,百度智能云提供了充沛的资源保障,以应对游戏可能遇到的QPS等高并发挑战).
NetEase Fuxi and Baidu AI Cloud used ERNIE large models to build the world's first in-game AI arena for Justice and deliver real-time intelligent NPC interaction (original: 网易伏羲联合百度智能云,基于文心大模型打造全球首个游戏AI竞技场,为《逆水寒》上亿玩家提供智能NPC实时交互体验).
The ERNIE character model can cover diverse settings and make stories develop more dynamically while keeping characters believable and fun (original: 文心人设大模型(ERNIE-Character-Fiction-8K)时空和场景优化能力…模型能够理解上下文,并主动推动多样化剧情发展).
Traditional NPCs are limited by predefined text, whereas the intelligent-agent approach gives every NPC more realistic personality and thought (original: 传统NPC受限于规定剧情文本的局限性…智能体最大的意义,莫过于赋予每个NPC独一无二的思想、性格).
Transparency Evidence & methodology 71/100 reference value · 87% confidence · 1 source

Review how this reference was scored and trace each normalized claim back to its documented source.

01

Scoring methodology

How scores are calculated

Customer Reference Value measures usefulness as a customer reference. Evidence Confidence measures the quality and traceability of available evidence, not independent verification.

Customer reference value71/100
  • Problem clarity20/20
    Measures whether the customer's operational problem is specific enough to compare.
  • Implementation depth25/25
    Rewards named components, data flow, integrations, and architecture detail.
  • Outcome specificity16/20
    Measures whether the reported benefit is concrete rather than a generic claim.
  • Quantified impact0/25
    0 metrics describe measurable impact.
  • Deployment scale10/10
    Rewards evidence of a live or scaled deployment and meaningful integration scope.
Evidence confidence87%
  • Source authority20/25
    The main evidence is an official vendor-published customer story.
  • Claim traceability21/25
    10 of 12 key fields link to a named source.
  • Direct evidence20/20
    4 direct source quotes retained.
  • Date provenance10/10
    The publication date comes from the source page or publisher.
  • Source transparency8/10
    1 source reference document publisher, URL, authority, retrieval, and supported fields.
  • Implementation specificity8/10
    Rewards source-backed components, data flow, quantified impact, and a specific outcome.
02

Traceability

Sources & provenance

Field coverage shows which facts each source supports. Dates are shown only when documented by the source.

  1. Baidu AI Cloud

    NetEase Fuxi customer story

    First-party
    https://cloud.baidu.com/customer/case/wangyi.html
    Source type
    Vendor-published customer story
    Published
    Jun 17, 2025
    Supports
    • Customer
    • Business problem
    • Solution
    • Products
    • Solution components
    • Data flow
    • Integrations
    • technical area
    • use case category
    • Business outcome
    • outcome category
    • Metrics
    • Architecture
    • evidence quotes
    • Publication date

Date evidence

  • Published on source page Jun 17, 2025 · Baidu AI Cloud · First-party
    Open date source
Not separately sourced Industry, Region, Company size. Treat these normalized attributes as lower-confidence until a dedicated source is added.