Tencent Calls Hy4 One of the World's Leading Open-Source Models. There Are No Weights Yet, and Artificial Analysis Scores It 35, Eleven Points Behind Kimi K3.1 Pro.
Tencent released Hy4 on November 4 with text and image input, a 1 million token context, and the same price as Hy4 Preview, about $0.90 in and $2.70 out per million tokens, with 50% off on Tencent Cloud through December 31. The press release calls it open source; the weights and system cards are due in the coming days, and as of November 6 Hugging Face still shows only the August Hy4 Preview. Artificial Analysis lists Hy4 as proprietary at 35 on Intelligence Index v5.0.1, 43rd of 106, using 160 million output tokens against a 75 million median. Tencent's finance panel ranks Hy4 first; Tencent ran the panel. We read the release and AA's model page.
By T.W. Ghost
Tencent released Hy4 today, November 4, 2026, under a press-release headline that puts it among "the world's leading open-source AI models." The model takes text and image input, has a 1 million token context, and costs the same as Hy4 Preview did: about $0.90 per million input tokens, $2.70 output, and $0.045 cached, or 6, 18, and 0.3 yuan. Tencent Cloud's TokenHub takes 50% off through December 31, and the WorkBuddy and CodeBuddy apps run it free from November 4 to 15.
Two words in the headline need a date. "Open-source" is a plan: the release says weights and system cards arrive "in the coming days," and as of the morning of November 6 Hugging Face still shows only the August Hy4 Preview and its FP8 variant, with no repository for the model Tencent just shipped. "Leading" has an independent number against it: Artificial Analysis lists Hy4 as proprietary at 35 on Intelligence Index v5.0.1, 43rd of 106 models, above the median of 30 and eleven points under Kimi K3.1 Pro at 46. It got there using 160 million output tokens, more than twice the 75 million median, which means the per-token price understates what a task costs. We read the release, AA's model page, and AA's changelog. Here is what held up.
What You Are Calling
| Tencent Hy4 | |
|---|---|
| Released | November 4, 2026 (AA dates it November 3) |
| Input | Text and image; 1M context |
| Price | About $0.90 in / $2.70 out / $0.045 cached per 1M; 6 / 18 / 0.3 yuan; same as Preview |
| Promotions | 50% off on Tencent Cloud TokenHub through December 31; WorkBuddy and CodeBuddy free November 4 to 15 |
| Weights | "In the coming days"; none on Hugging Face as of November 6 |
| Size | 770B total, 49B active is the Preview's size, per press; Tencent did not restate it |
| AA Intelligence Index v5.0.1 | 35, 43rd of 106; listed proprietary |
| AA output tokens on the Index | 160M, against a 75M median |
| Where | OpenRouter, Hermes, OpenCode (vendor) |
| Next | Hy4 Flash "soon" (vendor) |
Open is a promise until there is a repository with files and a license. This is the same shape as Kimi K3.1 Pro's October launch, where the weights arrived later than the headline. Until Hugging Face has Hy4, it is an API model.
Thirty-five is mid-pack, and verbose. Above the median on the new scale, well under the open-weight leaders, and burning twice the median output to get there. AA has not published a cost per task yet; when it does, expect it to be higher than the token price suggests.
The parameter count is borrowed. The 770 billion total and 49 billion active figures describe the Preview. Tencent's release does not give a size for Hy4.
The Benchmarks, Sorted by Who Reported Them
| Claim | Number | Who |
|---|---|---|
| Intelligence Index v5.0.1 | 35, 43rd of 106 | Independent (Artificial Analysis) |
| Kimi K3.1 Pro on the same index | 46 | Independent (AA changelog) |
| Finance blind evaluation | Hy4 3.04; DeepSeek V4.1 Flash 1017 2.97; Kimi K3.1 Pro 2.68; GLM-5.3 2.68 | Vendor (Tencent's own panel and tasks) |
| Hallucination rate | 12.95% down to 7.2% | Vendor |
A finance panel where the host finishes first is a vendor result, and we print it as one. The independent board is the one with Hy4 in 43rd place.
Verified vs Unconfirmed: The Scorecard
| Claim | Verdict |
|---|---|
| Released Nov 4; text and image input; 1M context; price and promotions | Vendor (Tencent release) |
| Weights and system cards "in the coming days" | Vendor; not released as of Nov 6 |
| AA Index v5.0.1 35, 43rd of 106; proprietary; 160M output tokens vs 75M median | Verified (Artificial Analysis) |
| Kimi K3.1 Pro 46 | Verified (AA changelog) |
| Finance panel ranking; hallucination reduction | Vendor |
| 770B total, 49B active | Press, and describes the Preview |
| Hy4 Flash timing | Vendor, "soon" |
Who Should Use It
Use Hy4 as what it is today: an API model priced in DeepSeek territory with a discount through year end, on an independent score that is respectable and not leading.
Skip the open-source framing until a repository exists, and check output volume on your own tasks before trusting the per-token price.
The frontier is elsewhere: among the open models that have shipped weights this week, Kimi K3.1 Pro finally has a repository and a license; among the ones that have not, Hy4 has company in Inkling 2 Small. Our model comparison has the current standings side by side.
Sources
- Tencent: Tencent launches Hy4, joining the ranks of the world's leading open-source AI models
- Artificial Analysis: Tencent Hy4 model page
- Artificial Analysis changelog
*Not sure whether a promised open model or a shipped one belongs in your plan? Take the free 2-minute quiz and get matched. Then read about the open model that kept its November 4 date: Kimi K3.1 Pro.*