█

AI/TLDR

Tencent · 2026-11-16 · major

Hunyuan Hy4 Flash — Tencent's cheap 1M-context model for everyday work

In short

Tencent released Hunyuan Hy4 Flash, a 224B-parameter MoE with 12B active and a 1M-token context. It costs $0.070 per million input tokens, about a twelfth of the full Hy4. Open weights are planned but not out yet.

Tencent · 2026-11-16 · 4 sources checked

Tencent chart of score against cost per task: Hy4 Flash sits at the lowest cost next to Hy4, DeepSeek and Qwen models
Tencent

Tencent's Flash-tier Hy4: a small-active MoE with a 1M-token context, built to make everyday agent work cheap and fast.

Key specs

Active params12B
Price (input)$0.070 / 1M tokens

Quick facts

MakerTencent Hy Team
Parameters224B total (64B Engram), 12B active
Context windowUp to 1M tokens
Price (output)$0.281 / 1M tokens
Price (cached input)$0.003 / 1M tokens
AvailabilityTencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima, Marvis
Open weightsPlanned, no date given

Pricing

Input$0.070 / 1M tokens
Output$0.281 / 1M tokens
Cached input$0.003 / 1M tokens
source ↗

What is it?

Hunyuan Hy4 Flash is a new low-cost model in Tencent's Hy4 family, released on November 16, 2026, twelve days after the full Hy4. It takes multimodal input, reads up to 1 million tokens, and is aimed at everyday work such as deep research, document processing, slides, simple front-end pages and computer use.

How does it work?

The model is a Mixture-of-Experts with 224B total parameters, 64B of them Engram memory parameters, and only 12B active per token. Tencent combines hybrid attention, LatentMoE and Engram to shrink the KV cache to 1/6 of Hy4's, trains with reinforcement learning to halve output tokens per task, and serves it with 4-bit quantization and megakernels at over 150 tokens per second.

Why does it matter?

Cost per task is the main pitch. On Tencent's internal WorkIndex benchmark, Hy4 Flash scores slightly above the Hy4 preview at 1/13 of its per-task cost and 1/20 of Hy4's. In a blind test with 85 WorkBuddy users, Tencent says satisfaction came close to Claude Haiku 5.5. These are company-run numbers, not independent results.

Who is it for?

developers running high-volume agents and document workflows

Frequently asked questions

Is Hunyuan Hy4 Flash open source?
Not yet. Hunyuan Hy4 Flash launched as a hosted model on Tencent Cloud TokenHub and inside Tencent apps such as WorkBuddy and Yuanbao. Tencent says it plans to open-source Hy4 Flash with frequent updates, but it has not given a date or a license, and OpenRouter support is listed as coming soon.
When should I use Hy4 Flash instead of the full Hy4?
Tencent recommends Hunyuan Hy4 Flash for everyday tasks such as deep research, document processing, PowerPoint generation, simple front-end displays and computer use. For complex, long-horizon software engineering Tencent still recommends the full Hy4, which scores higher in its own charts on both the internal WorkIndex and open benchmarks.
Can I try Hy4 Flash for free?
Yes, for a limited time. Tencent says Hunyuan Hy4 Flash is free to use in WorkBuddy until November 30, 2026, and Tencent Cloud TokenHub sells API access at 50% off the standard price over the same period. After that, the list price is $0.070 input and $0.281 output per million tokens.

Try it

Model id hy4-flash on Tencent Cloud TokenHub (50% off until November 30, 2026)

Sources · 4 outlets

Tags

  • tencent
  • hunyuan
  • hy4
  • hy4-flash
  • moe
  • long-context
  • multimodal
  • low-cost
  • china
  • llm

← All releases · Learn AI