Tencent released Hy4, its latest flagship model, on November 4, and said it would publish the weights and system card “in the coming days” (Tencent announcement). Two days later, Chutes (Bittensor subnet 64) said it was a launch partner with Tencent Hunyuan and that Hy4 was live on the same OpenAI-compatible API its users already call (@chutes_ai; Chutes model page).
On Chutes, Hy4 is listed as tencent/Hy4 at $0.525 per million input tokens, $1.5625 per million output tokens, and $0.025 per million cached input tokens (Chutes). That is about 42% below Tencent’s published standard API list. It is not below Tencent Cloud’s current launch discount.
This is a useful follow-up to our Nov 7 piece on Kimi K3.1 Pro. That story was about an open-weight model that independent operators could actually download and serve. Hy4 is different, and the difference matters.
What Tencent says Hy4 is
From Tencent’s Nov 4 announcement:
- Hy4 is Tencent’s “latest-generation large language model,” positioned as a step up from Hy4 Preview.
- It adds visual understanding and computer use for the first time in the Hy line: images, screenshots, charts, command-line interfaces, and graphical interfaces, with additional training on professional software such as Blender and CAD.
- Tencent reports its own evaluations, including a blinded financial-task study in which Hy4 scored 3.04 against 2.97 for DeepSeek-V4.1-Flash-1017 and 2.68 for both Kimi K3.1 Pro and GLM-5.3. Those are Tencent’s figures, on Tencent’s tasks.
- Standard API pricing is given as approximately $0.90 per million input tokens, $2.70 per million output tokens, and $0.045 per million cached input tokens — unchanged from Hy4 Preview, according to Tencent.
- Hy4 is available in Tencent products and, Tencent says, via third-party platforms including OpenRouter, Hermes, and OpenCode. From November 4 to 15, WorkBuddy and CodeBuddy are free. On Tencent Cloud TokenHub, Hy4 is 50% off standard pricing through December 31, 2026.
Tencent’s launch post does not publish a parameter count or a context window for the official Hy4 release, and it does not name Chutes among those third-party API platforms. The 770B-total / 49B-active Mixture-of-Experts figures and the 1,048,576-token context window are documented for Hy4 Preview, not in the official Hy4 announcement (Hugging Face, Hy4 preview). OpenRouter’s listing for official Hy4, served by Tencent Cloud, also shows a 1,048,576-token context window (OpenRouter). We treat preview architecture details as preview-only until Tencent publishes the Hy4 system card.
The prices, side by side
Checked this morning (Nov 10, America/New_York):
| Where | Input / 1M | Output / 1M | Cached input / 1M | Notes |
|---|---|---|---|---|
| Tencent (standard list) | ~$0.90 | ~$2.70 | ~$0.045 | Official list; TokenHub is 50% off this through Dec 31 (Tencent) |
| OpenRouter (Tencent Cloud, current promo) | $0.417 | $1.251 | $0.021 | Labeled 50% off; implied list $0.834 / $2.501 / $0.042 (OpenRouter) |
| Chutes (SN64) | $0.525 | $1.5625 | $0.025 | Model ID tencent/Hy4 (Chutes) |
Three takeaways:
- Chutes is cheaper than Tencent’s standard list, by about 42% on input and output. That is a real discount to the published sticker price.
- Chutes is not the cheapest route right now. Tencent Cloud TokenHub is 50% off through year-end, and OpenRouter’s Tencent Cloud endpoint is listed at $0.417 / $1.251 — below Chutes. Anyone comparing “decentralized vs. official” prices has to use the promotional official price, not just the standard list, until that discount expires.
- A list price is not cost per accepted result. As we argued in our inference-cost piece, retries, latency, and failures can wipe out a headline discount. OpenRouter currently shows Tencent Cloud at about 57 tokens/s throughput and 3.29 s P50 latency on its public board (OpenRouter). We did not find comparable public latency or success-rate figures on the Chutes Hy4 page.
What “launch partner” does and does not mean
On Nov 6, when the model went live, Chutes called itself a launch partner with Tencent Hunyuan (@chutes_ai; Taosis transcript). The next day the Opentensor Foundation said the subnet had partnered with Tencent Cloud and repeated the launch-partner claim (Taosis transcript of @opentensor, Nov 7). We have not seen Tencent itself name Chutes in the official launch post.
That leaves an important open question: is Chutes serving Hy4 on subnet miners, or routing to Tencent Cloud? The public record does not settle it. Official Hy4 weights were promised “in the coming days” after Nov 4. As of this morning, Hugging Face still shows public repositories only for Hy4 preview and Hy4 preview-FP8, both dated late August, not a public tencent/Hy4 weights release (Hugging Face, Hy4 preview). Without public official weights, independent operators cannot simply download Hy4 and host it the way they did with Kimi K3.1 Pro.
So the conservative reading is: Chutes has a day-one API for a flagship model, at a discount to Tencent’s standard list. That is a distribution win. It is not yet evidence that Bittensor miners are the ones running the official Hy4 checkpoint.
Why this still matters for Bittensor
Even as a distribution deal, the listing fits the pattern we have been tracking. Inference subnets are trying to be the place developers already have a key when a new model ships. If official Hy4 weights do come out, the Kimi playbook — many operators, price competition, optional TEE routes — can start. Until then, “Hy4 on Chutes” should not be counted as the same phenomenon.
The payment side is unchanged from our Nov 9 Gamma update. Gamma’s code is in runtime 488, but minting still depends on a network-wide flag, a per-subnet flag, and a live TAO/USD oracle, all of which were still off as of that review. We have not seen a public announcement since then that root or a subnet owner has flipped those switches. Hy4 traffic on Chutes, like Kimi traffic, would still settle off-chain until an inference subnet actually enables Gamma.
What to watch
- Official Hy4 weights and system card. Tencent said they would follow the Nov 4 launch. Until they are public, treat hosting claims with caution.
- How Chutes is serving the model. A clear statement — or evidence — that SN64 miners are running official Hy4, versus a Tencent Cloud API route, would change how to read this partnership.
- Pricing after December 31. TokenHub’s 50% launch discount is scheduled to end then. That is when Chutes’ discount to official list becomes a cleaner comparison.
- Public reliability data from Chutes. Latency, throughput, and success rates, not just the $0.525 / $1.5625 sticker.
- Gamma enablement on any inference subnet that is already selling model access.
Research and commentary only, not financial advice. Prices, listings and protocol flags are from the linked sources as of Nov 10, 2026 (America/New_York) and change often.