iFlytek Spark 讯飞星火
iFlytek's LLM line: free assistant, domestic-compute X2.5, small open models
Overview
iFlytek Spark (讯飞星火) is iFlytek's large-model line: a free consumer assistant, the 293B-parameter Spark X2 and X2.5 models that iFlytek says were trained on domestic Chinese compute, open-weight X2.5-4B and 1.7B models with a one-million-token context, and a yuan-priced API.
Pricing
Pricing shown for reference only. These figures reflect RECATOOLS research as of 14 Oct 2026 and may be out of date or incomplete.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.
- Deep research, writing, translation and reading
- Image generation
- Login by mainland phone number or WeChat
- Cache hits ¥0.24 per M tokens
- 256K context
- 200+ languages
- Maths, reasoning, language and agent tasks
- Agent and coding tasks
- 1M-token context
- 200+ languages
- Runs on vLLM, llama.cpp and Ollama
Use cases
What you can produce with iFlytek Spark 讯飞星火
- Chinese-language writing, translation and document reading in a free assistant
- API access to the 293B Spark X2.5 for coding and agent tasks
- On-device or self-hosted generation with the open X2.5-4B and 1.7B models
- Long inputs of up to a million tokens on the small open models
- Speech recognition across Mandarin and 202 dialects through iFlytek's speech models
- Industry deployments in education, healthcare and car cockpits through iFlytek sales
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。
Spark is a credible, cheaply priced Chinese model line with real strengths in speech, education and healthcare, and the September 2026 open release of X2.5-4B and 1.7B gives developers small Apache 2.0 models with a one-million-token context. The X2.5 API, at ¥1.60 in and ¥6.00 out per million tokens on a half-price offer, is inexpensive.
Its benchmark claims are iFlytek's own, the assistant needs a mainland phone or WeChat login, and billing and data sit in China. A good choice inside China's education and voice ecosystems; outside them, the open small models are the most useful part.
What we found
Spark (讯飞星火) is the large-model line of iFlytek, the Hefei speech-recognition company, and it covers a consumer assistant on web and mobile, an API platform and a set of industry models for education, healthcare and car cockpits. The assistant is free to use with a mainland phone number or WeChat login; no paid consumer plan appeared on the pages read.
The model line moved twice in 2026. Spark X2, launched on 11 February, is a 293-billion-parameter mixture-of-experts model that iFlytek says was trained on domestic Chinese computing power and, once quantised, runs on a single Huawei Ascend server. iFlytek claims it matches the best international models in maths, reasoning and agent tasks, and that inference runs 50% faster than the previous X1.5. Those are vendor claims on vendor test sets. Spark X2.5 followed on 7 September, keeping the 293-billion size with about 30 billion active, a 256K context, more than 200 languages and a stated focus on coding and agents.
A week earlier, iFlytek's subsidiary 词元星火 open-sourced two small models, Spark X2.5-4B and 1.7B, under Apache 2.0 on Hugging Face and GitHub. Both accept up to a million tokens of context, which the Science and Technology Daily described as a first for on-device models, and both run on Nvidia, Huawei and other hardware through vLLM, llama.cpp or Ollama.
API prices are low and published in yuan. Spark X2.5 is listed at ¥1.60 per million input tokens and ¥6.00 per million output, marked as a limited-time half-price offer; X2 costs ¥3 per million and X2-Flash ¥1 to ¥2. The 1.7B model's API is free and the 4B's free for a limited time. The same platform also resells GLM, Kimi and DeepSeek models.
For ASEAN teams the practical limits are the Chinese-first interface, mainland login, yuan billing and data held in China. Spark is most compelling where iFlytek's speech and education strengths matter, or for the small open models.
Compiled by RECATOOLS from public sources: vendor documentation and pricing, code repositories and published reporting.
Frequently asked questions
About this listing
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.
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 →
Alternatives to iFlytek Spark 讯飞星火
More in LLMs & Chat
