Summary:

Tencent’s lower-priced Hy4 flash shares Hy4’s documented context and output limits. Compare list prices, discount deadlines and the checks that matter before routing real work.

Tencent’s Hy4 flash is now listed on TokenHub, adding a lower-priced option alongside the full Hy4 model. The November 16 documentation identifies it as hy4-flash, with image understanding, tool calling and a one-million-token context window.

For developers choosing between the two, the immediate difference is the token price. The published capability list is similar, so a useful pilot should measure whether the cheaper model can complete the same work reliably.

Hy4 flash versus Hy4: published API prices

These are Tencent Cloud’s listed USD prices per million tokens, checked November 16, 2026, before temporary discounts.

Billing itemHy4 flashHy4
Input$0.070$0.834
Output$0.281$2.501
Cached input$0.003$0.042

Tencent is billing Hy4 flash at half the list price until noon on November 30, Beijing time. Hy4 has a separate half-price period ending at 23:59:59 on December 31, Beijing time. The discounts apply to usage incurred during their respective periods.

For budgeting, keep the discounted launch bill separate from the steady-state estimate. Token prices alone also miss retries, longer answers and review time: compare the total cost of an accepted result.

What the documented limits support

TokenHub lists the same limits for Hy4 flash and Hy4: a 1M-token context window, up to 960k input tokens and up to 64k output tokens. It lists deep reasoning with preserved thinking, structured output, function calling, caching and image understanding for both. Those are service limits and feature declarations, not a claim of equal quality.

The model is distinct from the full Hy4 release covered earlier this month. Select the exact model identifier rather than treating “Hy4” and “Hy4 flash” as interchangeable names.

Integration checks before switching traffic

Tencent’s API overview lists native support for OpenAI Chat Completions, OpenAI Responses and Anthropic Messages for hy4-flash. The corresponding model service must first be activated in the console; an unactivated service returns a 402 error. Existing compatible clients can use TokenHub, but capability-specific parameters still need checking.

Start with a small set of representative tasks: extract fields from a document, interpret a chart and complete a tool-assisted workflow. Keep prompts and acceptance criteria fixed, record failed or retried tasks, and measure latency alongside cost. Check long-input examples separately from short requests. This makes the comparison useful for your workload instead of assuming that a matching feature checklist guarantees matching results.

Sources and image credit