Hy4's pricing story is simple: list prices are identical to the preview — more capability, same price — and the 50% discount ends Dec 31. The real pricing question isn't the unit price but the per-task cost: AA measured about $0.78 per task, roughly one-tenth of the same-scoring Grok; but Hy4 is very verbose, generating roughly twice the median output volume, so real bills have to be calculated per task, not per token.
Account one is the list price. TokenHub: ¥6 in, ¥18 out, ¥0.3 cache hits per 1M tokens, the same in Guangzhou and Singapore; international $0.834 / $2.501 / $0.042. OpenRouter charges the same. The RMB and USD prices are two separate price lists — the dollar figures are not a conversion promise for the RMB ones.
Account two is the discount price. From launch through Dec 31, 2026, 23:59 Beijing time, billing is at 50% of list: ¥3 / ¥9 / ¥0.15, international $0.417 / $1.2505 / $0.021. OpenRouter's live quotes already show the discounted price. The window has less than seven weeks left, and list prices resume after — build long-term cost models on list prices, not discounted ones.
Account three is the task cost. Artificial Analysis' independent measurement gives about $0.78 per index task, roughly one-tenth of the same-scoring Grok 4.7 (xhigh) and one-thirteenth of Qwen3.8 Max (per Tencent's citation of the AA comparison). But AA also recorded Hy4 as "very verbose": 160M output tokens generated during the evaluation, roughly twice the median. Output tokens cost three times input tokens, so verbosity pushes bills up directly — factor it in for long tasks with long outputs. The good news is caching: OpenRouter reports a cache-hit rate of about 93%, so long-context tasks with repeated prefixes may bill well below list price in practice.
| Channel | Model ID / entry | Price | Best for |
|---|---|---|---|
| Tencent Cloud TokenHub | hy4 | ¥6 / ¥18 (50% off through Dec 31) | Developers in China; OpenAI-compatible API; third-party models available for comparison |
| OpenRouter | tencent/hy4 | $0.417 / $1.2505 (discounted) | International developers; supports reasoning, tools, JSON schema |
| WorkBuddy / CodeBuddy | In-product | Free quota, Nov 4–15 | Zero-code trials of long-horizon agents and office collaboration |
| Yuanbao / ima / WeCom Dayuan | In-product | In-product pricing | Direct consumer and office use |
| Hermes / OpenCode | Platform API | Platform pricing | Coding-agent workflows |
OpenRouter's live operating metrics for reference: P50 throughput about 63 tok/s, P50 time to first token about 2.9s, 99.87% availability. A listing site also quoted Novita AI at $0.42 / $1.25 (checked Nov 12, not verified by this site).
WorkBuddy and CodeBuddy offered a free quota Nov 4–15, later extended to Nov 15, 23:59. Users hit limits on day one; on Nov 5 Tencent expanded the Hy4 inference cluster tenfold and raised the free quota. Limits can still apply at peak times, and the official advice is to switch to Hy3 when that happens. The free quota is for trials, not production — production usage goes through TokenHub or OpenRouter paid APIs.
The full-release weights are unpublished, so self-hosting currently means the preview weights: BF16 about 1.56TB, FP8 about 814GB. Running FP8 needs roughly 500GB-class VRAM hardware (EveryInfra's figure, Nov 11); one aggregator claims a self-hosting floor of 16× B200 or 8× B300, but that has a single source and is unverified — this page does not present it as fact. The full release's weight format, quantization variants and hardware requirements can only be confirmed once the weights ship.
Reference: Tencent Cloud Developer Community — Hunyuan Hy4: Advantages and Application Scenarios (rewritten and organized by ModelBrief; prices per the official pages and checkout)