PROVIDER - Spark cognitive models with a speech ecosystem

iFlytek Spark

iFlytek Spark serves Spark Max and Spark tiers with speech recognition and synthesis under 1 account, supporting tool calling and knowledge bases through 2 access styles in WebSocket and HTTP plus education, office and support solutions; scattered docs and commercial enablement are the gaps.

AI Platform China 2026.11.19
4pts

Detailed Notes

Pros 5 items

Exposes Spark cognitive model APIs with 2 or more tiers such as Spark Max and Spark
Combines iFlytek speech technology so recognition and synthesis run under 1 account
Supports tool calling and knowledge bases to build 1 complete agent pipeline
Provides 2 access styles, WebSocket and HTTP, with WebSocket suited to realtime dialogue
Ships industry solutions across 3 scenarios: education, office and customer service

Cons 4 items

Documentation is large but scattered, raising search cost
Some capabilities require enabling subsystems and commercial approval
Model release cadence is gated by compliance review
Overseas nodes and English material are limited

Service Overview

iFlytek Spark is iFlytek's cognitive model platform, combining language models with speech technology in one open platform.

In the AI platform category it is the speech-rooted model platform: products needing recognition, synthesis and dialogue together can be built inside one account structure.

Key Strengths

  • Model tiers: Spark Max through Spark
  • Speech integration: recognition and synthesis together
  • Agent pipeline: tools plus knowledge bases
  • Two access styles: WebSocket and HTTP
  • Industry solutions: education, office, support

Product Ecosystem

Platform capabilities

Cognitive model APIs, speech recognition and synthesis, knowledge bases and tool calling, realtime dialogue access, industry customisation and usage management.

Integrations

Provides WebSocket and HTTP endpoints with multi-language SDKs, and composes with other iFlytek AI services.

Typical scenarios

Voice support and call centres, education and office assistants, meeting transcription and enterprise knowledge Q&A.

Limitations

  • Documentation structure: large but scattered
  • Enablement: some features need sales steps
  • Compliance cadence: releases gated by review
  • Overseas coverage: fewer nodes and English docs

Use Cases

  • Speech-integrated products (★★★★★): one account for both
  • Education and office scenarios (★★★★☆): mature vertical solutions
  • Realtime dialogue (★★★★☆): WebSocket support
  • Overseas multilingual business (★★☆☆☆): choose alternatives

Pricing

Item Price Notes
Usage based Per token or call Varies by capability
Resource packs Prepaid bundles Cheaper for steady usage
Industry solutions Contact sales Implementation and support included

FAQ

Which model tiers exist?

Spark Max and Spark among several tiers; see the official docs for the current list.

How is speech billed?

Recognition and synthesis are typically billed by call volume or duration.

Which interfaces are available?

2 access styles, WebSocket and HTTP, with WebSocket preferred for realtime.

Is there a free allowance?

The open platform usually offers trial credit after identity verification and capability enablement.

Code Examples

import requests, os

resp = requests.post(
    "https://spark-api-open.xf-yun.com/v1/chat/completions",
    headers={"Authorization": f"Bearer {os.environ['SPARK_API_KEY']}"},
    json={
        "model": "generalv4",
        "messages": [{"role": "user", "content": "Summarise 3 decisions from these minutes"}],
    },
    timeout=60,
)
print(resp.json()["choices"][0]["message"]["content"])
# List available models
curl https://spark-api-open.xf-yun.com/v1/models \
  -H "Authorization: Bearer $SPARK_API_KEY"

Community & Resources

Visit iFlytek Spark Website → https://xinghuo.xfyun.cn/
Visit Website ↗