Tencent Hy4 Flash vs Hy4: Cheap AI Compared

3D render of two glowing AI model cubes on scales, with the words "CHEAPER, SAME WORK?"

Tencent has launched a cheaper version of its flagship AI model: Hy4 Flash costs about one twelfth as much per input token. Launched on 16 November 2026 on Tencent Cloud TokenHub, it is aimed at everyday office work — writing, editing, spreadsheets and documents — rather than the hardest reasoning tasks. Can a model that costs this little actually do the work?

Here is what each one costs, what each one does, and where the claims need a second look.

Hy4 Flash: the numbers

Hy4 Flash is a 224-billion-parameter mixture-of-experts model that activates only 12 billion parameters per token, including 64 billion Engram parameters. It has a 1,048,576-token context window and native multimodality, meaning it can take image input as well as text. It is self-distilled from the full Hy4 model.

On TokenHub's Singapore list, the price is $0.070 per million input tokens, $0.281 per million output and $0.003 per million cached input. On the mainland list the rates are ¥0.5, ¥2 and ¥0.02. A launch discount halves them until 30 November 2026, and WorkBuddy is free until the same date.

It is sold through TokenHub as hy4-flash, and Tencent says it plans to release the weights — but has given no date.

Hy4: the flagship it undercuts

The full Hy4 model went generally available on 4 November 2026: 770 billion parameters, 49 billion active per token, the same million-token context window, and vision and computer-use capabilities.

Its price is $0.834 per million input tokens, $2.501 per million output and $0.042 per million cached input — so Hy4 Flash is about 12 times cheaper on input and almost 9 times cheaper on output.

The gap is the point: Tencent is selling Hy4 Flash not as a better model, but as a cheaper one for work that does not need the flagship. Drafting a presentation does not need 49 billion active parameters — it needs a model good enough and cheap enough to run all day.

What Tencent claims it can do

Tencent's headline claim is that Hy4 Flash matches Claude Haiku 5.5 on everyday work, based on its own blind test of 259 tasks run on its own employees — PowerPoint, co-writing, document editing and table repair.

In that test, Hy4 Flash scored 2.79 out of 4 on satisfaction, level with Claude Haiku 5.5 at 2.81 and ahead of GLM-5.3-Flash at 2.53. Its cost per task was ¥0.26 against ¥4.73 for Haiku, and its median task time of 6.7 minutes was the shortest tested.

Those numbers need careful handling. The price is real and the blind test is real, but the comparison is Tencent's own arithmetic, run on its own employees on its own benchmark, WorkIndex. There is no independent benchmark, no model card and no licence published — so the claim is Tencent's, not a verified result.

Hy4 Flash vs GLM-5.3-Flash: the rival that is actually open

GLM-5.3-Flash, open-sourced on 26 August 2026, bills $0.15 per million input tokens, $0.50 per million output and $0.03 per million cached input. Tencent is roughly half the price on input and output, and ten times cheaper on cached input.

But GLM-5.3-Flash has something Hy4 Flash does not: it is MIT-licensed, its weights are on Hugging Face, and it scores 27.8 on the Artificial Analysis Intelligence Index. Hy4 Flash has no weights, no licence and no independent score. It is cheaper on paper and less verifiable in practice.

Which one should you pick

If you want the lowest price for office work and already use Tencent Cloud, Hy4 Flash is the cheapest of the three, and free WorkBuddy access until 30 November makes it easy to try. If you need to run a model on your own hardware, or want an independent benchmark first, GLM-5.3-Flash is the safer choice — open, downloadable and independently scored.

The real test is what happens after 30 November, when the discount ends, and whether Tencent open-sources the weights.

Key takeaways

  • Tencent launched Hy4 Flash on 16 November 2026 on TokenHub, priced at $0.070 per million input tokens and $0.281 per million output, with a 1-million-token context window.
  • Hy4 Flash is about 12 times cheaper than Hy4 on input and almost 9 times cheaper on output, with the same context window.
  • A 50 percent launch discount runs until 30 November 2026, and WorkBuddy is free until the same date.
  • Tencent's own blind test of 259 office tasks put Hy4 Flash level with Claude Haiku 5.5 on satisfaction, at about one eighteenth of the cost per task — not independently verified.
  • Against GLM-5.3-Flash, Hy4 Flash is roughly half the price on input and output, but GLM-5.3-Flash is MIT-licensed, with weights on Hugging Face and an independent score of 27.8.
  • Hy4 Flash has no model card, no independent benchmark and no released weights; Tencent plans to open-source it, undated.

Conclusion

Hy4 Flash is the cheapest way into a million-token model for office work, and Tencent's own test says it holds its own against Claude Haiku. But without weights, a model card or an independent benchmark, the claim rests on Tencent's word alone. Try it free in WorkBuddy before 30 November.

FAQ

What is Tencent Hy4 Flash?

Hy4 Flash is a low-cost AI model launched by Tencent on 16 November 2026, with 224 billion total parameters, 12 billion active per token, a 1-million-token context window and native multimodality. It is self-distilled from the flagship Hy4 model for everyday office work.

How much cheaper is Hy4 Flash than Hy4?

On TokenHub's Singapore list, Hy4 Flash costs $0.070 per million input tokens and $0.281 per million output, against $0.834 and $2.501 for Hy4 — about 12 times cheaper on input and almost 9 times on output. A 50 percent launch discount applies until 30 November 2026.

Is Hy4 Flash open source?

Not yet. Tencent says it plans to release Hy4 Flash as an open-weight model but has given no date, and there is no model card, licence or weights repository. GLM-5.3-Flash is MIT-licensed with weights on Hugging Face.

Sources

Researched with AI assistance and reviewed for accuracy.

Follow on WhatsApp

Follow Daily Tech & AI on WhatsApp for one tech and AI update every day: Daily Tech & AI WhatsApp channel.

 

Comments