AI News | 4 min read

Tencent Launches Hy4 Flash: 224B-Parameter Model at $0.07 Per Million Tokens

Tencent released Hy4 Flash, a 224-billion-parameter multimodal model priced at $0.07 per million tokens — 92% cheaper than full Hy4 — available immediately across its major platforms.

Hector Herrera
Hector Herrera
A newsroom featuring documents, contracts, related to Hy4 Flash: 224B-Parameter Model at $0.07 Per Million Tokens from an unusual angle or perspective
Why this matters Tencent released Hy4 Flash, a 224-billion-parameter multimodal model priced at $0.07 per million tokens — 92% cheaper than full Hy4 — available immediately across its major platforms.

Tencent Launches Hy4 Flash: 224B-Parameter Model at $0.07 Per Million Tokens

By Hector Herrera | November 16, 2026 | News


Tencent released Hy4 Flash on November 16, a multimodal AI model with 224 billion total parameters priced at $0.07 per million input tokens — roughly 92% cheaper than the full Hy4 model it distills from. It is immediately available across Tencent's major consumer and enterprise platforms, with a free tier running through November 30.

The price point positions Hy4 Flash as a direct challenge to commodity inference providers, targeting the high-volume, cost-sensitive workloads that make up the bulk of enterprise AI usage.

What Hy4 Flash Is

Hy4 Flash is a distillation of Tencent's full Hy4 model — meaning it was trained to replicate the behavior and reasoning of the larger system in a more efficient architecture. According to Unite.AI's coverage of the launch, the technical specifications are:

  • Total parameters: 224 billion
  • Active parameters per token: 12 billion (mixture-of-experts architecture)
  • Context window: 1,000,000 tokens
  • Input pricing: $0.070 per million tokens
  • Multimodal: Yes — natively handles text, images, and documents
  • Availability: Tencent Cloud API, WeCom, QQ, Yuanbao, WorkBuddy (free through November 30)

For comparison, full Hy4 — released in late October — was priced for enterprise contracts, not API pay-as-you-go. Flash brings that underlying capability to individual developers and small teams.

The Distillation Approach

Model distillation is the process of training a smaller, faster model to mimic the outputs of a larger one. The smaller model doesn't have all the parameters of the original, but it learns to produce similar results on the tasks it's optimized for.

The tradeoff: Flash will underperform full Hy4 on complex multi-step reasoning, but for everyday productivity tasks — summarizing documents, answering questions, drafting content, processing images — the gap narrows significantly and the cost difference is dramatic.

Tencent's 12-billion active parameters per token (despite 224B total in the MoE pool) is a sweet spot that keeps inference fast and affordable. This is a similar approach to what Google uses in Gemini Flash and Anthropic uses in Claude Haiku.

92% Cheaper: What That Means in Practice

The $0.07 per million token input price deserves context:

  • Full Hy4 was not publicly priced at launch, but comparable enterprise models from Chinese providers have run $0.60–$1.50 per million tokens.
  • GPT-5 Mini (OpenAI's cost-optimized tier) currently runs approximately $0.15 per million input tokens in the US market.
  • Claude Haiku (Anthropic's cost tier) is priced around $0.25 per million input tokens.

At $0.07, Hy4 Flash is competing at the absolute low end of the frontier-quality model market. For applications processing tens of millions of tokens per day — customer service, content moderation, enterprise search — that price difference is meaningful.

Platform Distribution

What sets this launch apart from most model releases is the breadth of immediate platform distribution. Hy4 Flash isn't only available as an API for developers; it's being rolled out inside:

  • WeCom — Tencent's enterprise communication platform (think Slack for Chinese enterprises)
  • QQ — one of China's largest consumer messaging platforms
  • Yuanbao — Tencent's AI assistant product
  • WorkBuddy — Tencent's AI productivity suite, free tier through November 30

This means Hy4 Flash will reach users who may not interact with it through an API at all — it's embedded in tools they already use. That's a distribution advantage that pure API providers don't have.

What This Means for the AI Model Market

The race to the bottom on model pricing continues to accelerate. Three dynamics are at work:

  1. Distillation makes frontier capability cheap. Training a small model on the outputs of a large one is significantly less expensive than training the large model in the first place, and the results have been converging.
  2. Chinese providers are pricing aggressively for global market share. Alibaba's Qwen4-Max launched on the same day at $3.27 per task; Tencent is undercutting even that.
  3. 1M-token context is now table stakes. Both Qwen4-Max and Hy4 Flash launch with million-token windows, pushing the industry expectation for what a production model offers.

For enterprise buyers, the practical question is no longer "can we afford frontier AI" — it's "which frontier model fits our compliance, latency, and reliability requirements."

What to Watch

Tencent hasn't published third-party benchmark results for Hy4 Flash as of launch. Independent evaluation on English-language tasks — not just Chinese-language benchmarks — will determine whether Western enterprises seriously consider it. The free WorkBuddy tier through November 30 is a trial driver; watch for adoption numbers from Tencent's Q4 earnings report.


Sources: Unite.AI — Tencent Launches Hy4 Flash as a Lower-Cost Multimodal Model

Key Takeaways

  • ✓ Active parameters per token:
  • ✓ Distillation makes frontier capability cheap.
  • ✓ Chinese providers are pricing aggressively for global market share.
  • ✓ 1M-token context is now table stakes.

Did this help you understand AI better?

Your feedback helps us write more useful content.

Hector Herrera

Written by

Hector Herrera

Hector Herrera is an AI systems architect in Houston and founder of Hex AI Systems. He designs and runs AI systems in production and writes daily about how AI is reshaping business, government and everyday life. 20+ years building for the web. Houston, TX.

More from Hector →

Get tomorrow's AI briefing

Join readers who start their day with NexChron. Free, daily, no spam.

More from NexChron