Tencent Releases Hy4 flash: 0.5 Yuan per Million Input Tokens, 1M Context
The model releases keep coming, and Tencent's Hunyuan team has added one. On November 16, Tencent Hunyuan Research published the architecture and pricing for Hy4 flash: 224 billion total parameters (including 64 billion Engram parameters), 12 billion activated, multimodal support, and a context window of 1M (one million) tokens. The number that drew the most attention was the price — 0.5 yuan per million input tokens and 2 yuan per million output tokens, which the company says is substantially lower than Hy4.
Tencent is positioning it as an everyday, high-value model. The release published the architecture, a price table, a cost comparison, and evaluations across several business scenarios — an invitation to work out for yourself whether it is worth using for routine tasks.
What makes this release interesting is not a contest over who has the biggest model. It is that a vendor is treating "good enough and affordable" as a serious direction in its own right.
Two: reading the parameters — 224B and 12B are different things
📖 Glossary
AI Box (also known as Agent Computer / Agent PC), is a dedicated local hardware device that runs AI Agents. Pre-installed with an AI agent management system, plug-and-play, running 24/7. Users can remotely command AI to work via Discord, Slack, Telegram, WhatsApp, and more.
The wording matters here.
224B is the total parameter count, of which 64 billion are Engram parameters — that is the scale of the model's inventory. 12B is the activated count, the part actually called on in a single request. This is sparse activation: build a large warehouse, then switch on only a small section when doing a job. Do not read 224B as the per-call cost; that would be a mistake.
Add multimodal support and a 1M context, and the positioning becomes clear. This is not chasing peak capability; it is aimed at everyday tasks being usable, affordable, and practical to run.
Three: the real story is the price table

The most striking thing about Hy4 flash is the price.
0.5 yuan per million input tokens, 2 yuan per million output tokens. Note the unit — it is per million tokens, not per request, so it is better read as a unit price. The company says this is substantially lower than Hy4, its own higher-tier model.
A major vendor pushing an everyday-tier model to this price sends a clear signal: the competition is shifting from "who is strongest" toward "who can ordinary people afford to keep using." The cost comparison and business-scenario evaluations in the release all point the same way — it is not trying to win every benchmark, but focusing on the tasks people actually repeat.
For most real uses — classification, summarization, extraction, rewriting, customer-service Q&A — a top-tier model is not always necessary. A cheap, capable model you can keep running often fits real demand better.
Four: three things worth keeping

One, check the unit before the price. 0.5 yuan is per million input tokens. Treat it as a unit price and estimate your own volume to see what a month really costs.
Two, separate inventory from activation. 224B is the warehouse, 12B is the per-call cost. Conflating them leads to the wrong cost calculation.
Three, the more you plan to keep it running, the more a local machine matters. The unit price has dropped, but dispatching jobs, receiving results, and storing files still need a machine that is always there.
Models like Hy4 flash have brought the price down. Inference still happens in the cloud, but the machine that calls them needs to stay awake. For ordinary users who want to put this to work, a natural companion is KAIHE AIBOX(铠盒AIBOX), a computer built specifically for agents and your local private AI assistant. OpenClaw and Hermes come installed from the factory, so it is ready on standby as soon as it is plugged in and online — no repeatedly rebuilding the environment. Inference goes to the cloud, while files, long-term memory, and call records stay local.
Unit prices will keep falling. The machine that is ready to work for you at any moment is the stable part.