iFlytek Spark 讯飞星火

iFlytek's flagship LLM — strong voice + Chinese capability

LLMs & Chat Freemium Has API
Researched · Published · Reviewed
RECATOOLS Score
6.2 / 10
Capability
7
Value for money
6
Ease of use
6
ASEAN readiness
4
API quality
6
Founded
2023
HQ
Hefei, Anhui, China
Users
Not publicly disclosed as a single figure; iFlytek reports its medical assistant has logged over 1
Launched
May 2023 (Spark V1; X1 deep-reasoning model Jan
Developer

Overview

iFlytek's flagship LLM family, strongest in Chinese voice interaction and vertical deployments across education, healthcare and automotive. Spark X2 (Feb 2026, 293B MoE) and X2-Flash (Apr 2026, 256K context) were trained entirely on Huawei Ascend compute — no Nvidia hardware.

Advertisement

Pricing

Pricing shown for reference only. These figures reflect RECATOOLS research as of 11 Jul 2026 and may be out of date or incomplete. This is not financial or purchasing advice — always confirm the current price on the provider’s official website before making any decision.

API pay-as-you-go
Usage-based
Token packages per model on the iFlytek Open Platform, billed in CNY.
  • Tiered token packages (bigger = cheaper)
  • Pro, Max/Ultra and X1 model tiers
  • Off-peak 0.8x rate since Jun 2026
Enterprise
Custom
Volume contracts and private deployment.
  • Custom volume pricing
  • Private/on-prem deployment options
  • Industry solutions via iFlytek sales

Use cases

Voice AI Education AI Automotive cockpit Real-time translation

What you can produce with iFlytek Spark 讯飞星火

  • Draft and refine a Chinese-language business proposal, marketing copy, or social media post with voice-input support and multimodal review
  • Generate a structured study plan or exam-preparation materials tailored to China's gaokao or university curriculum, with strong mathematics coverage
  • Obtain AI-assisted medical triage suggestions or prescription-risk screening (deployed across grassroots clinics in China)
  • Reduce enterprise contract review time using the Spark Contract Assistant's machine pre-review plus human recheck workflow
  • Produce multilingual meeting transcripts and summaries with Chinese-English bidirectional translation
  • Accelerate software R&D workflows using the Xinghuo Code model in finance and manufacturing contexts
  • Create advertising copy, presentation frameworks, or structured reports using Spark's AI writing assistant, with voice-dictation input and image-upload context
Advertisement

ASEAN Perspective

iFlytek Spark 讯飞星火 in Southeast Asia

在教育与车载语音两个垂直市场具备结构性优势。讯飞翻译机系列在东南亚旅游与商务场景已有可观存量。

RECATOOLS Verdict

Spark is the speech company's LLM, and it shows: Mandarin understanding, voice interaction and multimodal speech tasks are where iFlytek's two decades of speech-recognition heritage pay off. The 2026 X2 generation (a 293B-parameter MoE, with the 256K-context X2-Flash following in April) was trained entirely on Huawei Ascend compute, which makes Spark strategically important to China's domestic-silicon push even though leaderboards put DeepSeek, Qwen, GLM and Kimi ahead as generalists. It's a strong pick for education, healthcare and automotive-cockpit deployments inside China, with API access through iFlytek's platform and official SDKs on GitHub. For ASEAN or global use the constraints are familiar: Chinese-first product and accounts, thin English documentation, and mainland data flows that raise compliance questions.

Independent AI-assisted assessment by RECATOOLS.

What people say

Nobody puts Spark at the top of China's LLM leaderboards — DeepSeek, Qwen, GLM and Kimi own that conversation — but writing it off as an also-ran misses where iFlytek actually makes money: education devices and classroom tools, healthcare (iFlytek says its AI diagnosis assistant has logged over a billion assisted diagnoses across grassroots clinics), automotive cockpits and translation hardware.

On infrastructure, 2026 has been Spark's most interesting year. Spark X2 shipped in February as a 293B-parameter MoE model trained entirely on Huawei Ascend compute — the first flagship-line release with zero Nvidia hardware — followed on April 29 by X2-Flash, a 30B MoE with 256K context, also Ascend-trained and opened up via API. iFlytek claims support for 130+ languages and roughly 50% gains in deep reasoning over the prior generation, and an April 2026 "Spark Coding Plan" subscription chases the AI-coding wave. Companion releases include Spark Healthcare X2, which took all-A ratings in Shanghai's medical large-model evaluations, and the X2-VL multimodal model in June 2026.

User sentiment splits by use case. Voice input and Chinese comprehension draw consistent praise — two decades of speech-recognition heritage show — and the education and medical verticals rate well. General chat is more mixed: slower text generation than Doubao or Kimi, weaker image generation, occasional interface clunkiness and open-domain hallucinations. Western coverage is sparse and English documentation thin, though official SDKs sit on GitHub. Treat Spark as a vertical specialist with unusually strategic supply-chain positioning, not a frontier generalist.

Summary of public user & expert reviews, compiled by RECATOOLS.

About this listing

Researched on
Published on
Last reviewed

This entry was compiled from publicly available data including iFlytek Spark 讯飞星火's official website, press releases, documentation, and reputable third-party publications. RECATOOLS is not affiliated with iFlytek Spark 讯飞星火 unless explicitly stated.

Data accuracy

Third-party AI tools update their pricing, features, availability, and policies frequently. Information here may be outdated by the time you read this — we make reasonable efforts to keep listings current, but cannot guarantee absolute accuracy.

For the latest details, please refer to iFlytek Spark 讯飞星火 directly →

Spotted something out of date? Suggest an update →

Advertisement