Tencent released Hy4 on November 4, 2026, and it is selling at half price. Through December 31, 2026, the model bills at $0.417 per million input tokens, $0.021 per million cached input tokens, and $1.25 per million output tokens. The list price behind that is $0.834, $0.042, and $2.50. Tencent's announcement states the 50 percent promotion and its end date in the same sentence, which makes this one of the few launch discounts where the buyer knows exactly when the bill doubles.
Today there is one seller. Tencent serves Hy4 itself through TokenHub and OpenRouter, the final weights are promised but not yet published, and the six hosts that carry the earlier Hy4 Preview all charge the full list rate for the older build. This page covers the rates, the promotion window, who sells the model, how it compares, and what a buyer should do before January 1. For the model-by-model view, see the LLM API pricing comparison.
Tencent Hy4 official API pricing
USD per 1M tokens, from Tencent's endpoint on OpenRouter and the TokenHub international price list:
| Token type | Promotional price (through Dec 31, 2026) | List price |
|---|---|---|
| Input | $0.417 | $0.834 |
| Cached input | $0.021 | $0.042 |
| Output | $1.2505 | $2.501 |
The context window is 1,048,576 tokens, with a 65,536 token maximum output on Tencent's OpenRouter endpoint. The model takes text and image input. Neither Tencent page we read lists a long-context surcharge.
One note on the list price. Tencent's English announcement rounds it to approximately $0.90 input, $2.70 output, and $0.045 cached. TokenHub's international page and OpenRouter both show $0.834, $2.501, and $0.042, and OpenRouter flags the current rate as a 50 percent discount from exactly those figures. We use the listed numbers. Check your own invoice against them.
When the promotion ends and what it costs
Tencent says the 50 percent offer runs through December 31, 2026. For a production assistant doing 10 million input tokens and 1 million output tokens a day, uncached:
| Rate | Daily | Annual |
|---|---|---|
| Promotional price, through Dec 31 | $5.42 | $1,978 |
| List price, from Jan 1 | $10.84 | $3,957 |
Same workload, same model, $1,979 a year apart, and the date is on the calendar. A team that builds on Hy4 in November is planning around a price with seven weeks left on it. Caching softens the number but not the ratio: at a 60 percent cache hit rate the workload runs $3.04 a day on the promotion and $6.09 at list.
This is the case a forward contract is built for. A launch promotion is a temporary price, and a forward is how a buyer locks future capacity at a known rate before it goes back up. That is what the Mercatus Forward Market does, and how to lock in token prices walks through the mechanics.
We have watched this exact sequence once already this quarter. Mistral Large 4 launched October 6 at a 50 percent promotion with no stated end date, moved to its full list price around October 20, and reset again nine days later. Tencent has at least published the date.
Who sells Tencent Hy4
One company, for now. As of November 10, 2026, OpenRouter lists a single endpoint for Hy4, run by Tencent, at the promotional rate. Tencent also sells it directly through TokenHub, and says the API is reachable through third-party platforms including OpenRouter, Hermes, and OpenCode. Those are routes to Tencent's own serving, not independent hosts.
The weights are the open question. Tencent's announcement says it plans to release the Hy4 weights and system cards "in the coming days." As of November 10, Tencent's Hugging Face account carries only the Hy4 Preview repositories from August. Until the final weights are out, no third party can serve this build.
The Preview shows what the host market looks like once they are. Hy4 Preview shipped in late August under Apache 2.0, a license with no hosting restrictions, and six providers list it on OpenRouter: Tencent, DeepInfra, Novita, SiliconFlow, GMICloud, and AtlasCloud. All six charge exactly $0.834 and $2.501. That is a spread of 1x, the same pattern as Qwen, where hosts match the official rate and compete on speed. It also means Tencent's own promotion on the newer model is currently half the price of every host serving the older one. Why token prices differ covers why some open-weight markets spread out and others sit at parity.
How Hy4 compares
Official rates per 1M tokens, with the daily cost of the same 10M input and 1M output workload:
| Model | Input | Output | Daily cost |
|---|---|---|---|
| Tencent Hy4, promotion (through Dec 31) | $0.417 | $1.25 | $5.42 |
| Tencent Hy4, list | $0.834 | $2.50 | $10.84 |
| DeepSeek V4 Pro, off-peak | $0.66 | $1.98 | $8.58 |
| Mistral Large 4 | $0.70 | $2.10 | $9.10 |
| DeepSeek V4 Pro, peak | $1.32 | $3.96 | $17.16 |
| GLM-5.3 | $1.40 | $4.40 | $18.40 |
| Kimi K3.1 Pro | $2.00 | $10.00 | $30.00 |
On the promotion, Hy4 is the lowest-priced model in this set. At list it lands between DeepSeek V4 Pro off-peak and its peak rate, about 19 percent above Mistral Large 4, 40 percent below GLM-5.3 on input, and well under half of Kimi K3.1 Pro. Tencent's own launch evaluation names DeepSeek V4.1 Flash, Kimi K3.1 Pro, and GLM-5.3 as the comparison set, so the company is pricing against the same models a buyer would shortlist.
What Tencent Hy4 is
From Tencent's announcement and the public listings:
- Release: November 4, 2026, following Hy4 Preview in late August.
- Lineage: Hy4 Preview is a mixture-of-experts model with 770 billion total parameters and 49 billion active. Tencent has not yet published the system card for the final release.
- New in the final release: image understanding and computer use, the first Hy model with either.
- Context: 1M tokens.
- Tencent's reported gains over Preview: hallucination rate down from 12.95 percent to 7.2 percent, and a 3.04 score on an internal blind financial evaluation of 579 tasks, against 2.97 for DeepSeek V4.1 Flash and 2.68 for Kimi K3.1 Pro and GLM-5.3. These are Tencent's figures and have not been independently reproduced.
- Weights: promised, not yet released. The Preview weights are Apache 2.0; the license for the final weights has not been published.
- Coming: Tencent says a smaller Hy4 Flash is scheduled for release soon, with no price given.
What buyers should do before December 31
Budget at the list price. Any cost model for 2027 should use $0.834 and $2.50. The promotional rate is a seven-week trial window.
Use the window to test. Half price through year end is a cheap way to run evals and find out whether Hy4 earns the list rate on your workload.
Watch for the weights. If the final weights ship under the same Apache 2.0 terms as the Preview, hosts can list Hy4 without asking permission. The Preview market suggests they will match Tencent's list price, not undercut it, so do not plan on a third-party discount replacing the promotion.
Lock what you can. If the model fits and the volume is predictable, a forward contract fixes the rate past January 1 so the reset happens on your terms.
To track token prices across models and hosts, see the Mercatus Token Index.
Frequently asked questions
How much does the Tencent Hy4 API cost?
Through December 31, 2026: $0.417 per 1M input tokens, $0.021 per 1M cached input tokens, and $1.25 per 1M output tokens. The list price is $0.834, $0.042, and $2.50.
When does the Tencent Hy4 promotion end?
Tencent says the 50 percent offer runs through December 31, 2026. From January 1 the model bills at list price unless Tencent extends it.
Is Tencent Hy4 open weight?
Not yet for the final release. Tencent says it plans to publish the Hy4 weights and system cards in the coming days. As of November 10, 2026, only the Hy4 Preview weights are available, under Apache 2.0.
Who hosts Tencent Hy4?
Only Tencent as of November 10, 2026, through TokenHub and a single OpenRouter endpoint. Six providers host the earlier Hy4 Preview, all at $0.834 input and $2.501 output.
What is the difference between Hy4 and Hy4 Preview?
Hy4 is the official release of November 4, 2026. It adds image understanding and computer use, and Tencent reports a lower hallucination rate and stronger agentic performance than the Preview from August. The list price is the same.
What is the context window of Tencent Hy4?
1M tokens (1,048,576), with a 65,536 token maximum output on Tencent's OpenRouter endpoint.
Methodology
Release date, promotion terms and end date, availability, weights plan, new capabilities, and benchmark claims are from Tencent's Hy4 announcement of November 4, 2026. Promotional and list rates, cached rates, context window, maximum output, and the 50 percent discount flag are from OpenRouter's Tencent Hy4 endpoint data on November 10, 2026, and match the Singapore-region prices on Tencent Cloud's TokenHub page; the announcement's rounded figures of approximately $0.90, $2.70, and $0.045 are noted above. Hy4 Preview parameter counts, host list, and prices are from OpenRouter the same day. Weights availability and the Preview license are from Tencent's Hugging Face account, checked November 10, 2026. Mistral, DeepSeek, Z.ai, and Moonshot rates are from the linked Mercatus pages. Workload examples use stated token volumes at 365 days; recompute with your own mix. Last verified: 2026-11-10.