Tencent’s Hy4 flash costs $0.04 per million input tokens and keeps the 1M context
Tencent's smaller Hy4 flash is on OpenRouter at $0.04375 per million input tokens, with the flagship's 1M-token window, image input and tool calling.

Tencent’s Hy4 flash became available on OpenRouter on November 16. It is a cheaper variant of the Hy4 flagship, and it costs $0.04375 per million input tokens and $0.1756 per million output tokens while keeping a context window of about one million tokens.
A quarter of the parameters, the same limits
OpenRouter’s listing says Hy4 flash is served with roughly a quarter of the flagship’s parameter count, which it says brings lower latency and cost. Tencent, according to the listing, aims it at code assistance, agent pipelines, document and chart comprehension, and high-volume routine automation. The context and output limits match Hy4.

The listing gives these specifications:
- Context window: 1,048,576 tokens, with up to 65,536 completion tokens.
- Input: text and images. Output: text.
- Function calling through
toolsandtool_choice, plus structured outputs through a JSON schema inresponse_format. - Provider: Tencent Cloud only. OpenRouter says it forwards every request without routing between providers.
- Measured speed: a median latency of 1.28 seconds and 126 tokens per second.
- Region: EU in-region routing is not currently available.
The listed prices are $0.07 per million input tokens and $0.281 per million output tokens. A 38% discount brings them down to the figures above, and cached input reads cost $0.001875 per million tokens. OpenRouter does not say how long the discount lasts. Its own data shows buyers paying a weighted average of $0.03811 for input and $0.1389 for output, though that rests on very little traffic so far. The page lists about 636K tokens across its top apps, led by 569K from Claude Code.
What Tencent has not yet said about the model
The listing gives no parameter count, benchmark scores or architecture details for the flash model. A quarter of the flagship’s size would be roughly 190 billion parameters by simple arithmetic, based on the 770 billion figure reported for Hy4, but Tencent has not stated that number. All the performance claims on the page come from the listing itself and are not independently verified.
Tencent had signaled the model in advance. According to TechNode’s report on the November 4 Hy4 launch, Tencent planned a lower-cost Flash model but gave no details or timeline. The flagship added vision and computer use, and it is available through Tencent’s products and third-party platforms including OpenRouter. TechNode also reported that Tencent planned to release the weights and a system card shortly, without a date. Nothing on the flash page says whether it will get open weights.
Roughly a tenth of the flagship’s price
The comparison with Hy4 is the clearest measure of the saving. TechNode reports standard Hy4 API pricing of $0.834 per million input tokens and $2.501 per million output tokens, unchanged from the Hy4 preview, with a 50% discount on Tencent Cloud TokenHub through December 31. OpenRouter’s Hy4 page lists $0.417 and $1.251, which matches that halved rate. Against those figures, flash’s discounted input price is about 10 times lower, and its output price about 7 times lower. At list prices, the gap is roughly 12 times on input and 9 times on output. Both models are on discounts, so the comparison holds only while the promotions last.
Among other budget models on OpenRouter, the price sits low. Baidu’s ERNIE-5.5-Flash-Preview, which we covered when it landed on OpenRouter at $0.28 per million input tokens, costs about six times as much per input token. Meta’s Muse Ember, which showed up at $0.10 per million input tokens, is more than twice the price. Neither article contains a quality comparison with Hy4 flash, and price per token says nothing about how many tokens a task needs or how well the model does it.
Chinese labs have been cutting prices on small, long-context models in this way for weeks. Baidu’s flash model also arrived with a price and a promise of weights, as our report on ERNIE 5.5 Flash described. Tencent’s version adds image input and a million-token window at a lower price. Whether it is good enough for the agent and coding work it targets can only be settled by independent tests, which do not exist yet.
Who can use it now
Any developer with an OpenRouter key can call tencent/hy4-flash through the OpenAI-compatible API by changing the base URL and the model slug. The listing shows one provider, so a Tencent Cloud outage would leave no fallback. European teams with data-residency requirements should note the missing in-region routing. For high-volume, low-stakes jobs such as classification, document extraction or routine agent steps, the price makes it an inexpensive model to test. For anything where accuracy matters, Tencent’s benchmarks and the system card promised for the flagship are what to wait for.
Sign Up for the Daily AI Pulse
One email a day. All the stories that matter.
By signing up, you agree to our Terms of Use and acknowledge the data practices in our Privacy Policy. You may unsubscribe at any time.
Alex CarterAI News & Big Tech CorrespondentAlex Carter writes for Aiholics, keeping readers updated on the fast-paced world of AI and Big Tech.
You may also like

Gemini 4 Carbon opens to all developers at $2 per million input tokens

OpenAI’s gpt-6-rosalind costs $75 per million output tokens, above its GPT-6 flagship

Tencent’s Hy4 appears on OpenRouter at half the preview’s price
