Hy4 flash: Power-User Guide to Tencent’s Low-Cost API and Limits

Last updated: November 16, 2026

Hy4 flash is Tencent’s lowest-cost Hy model, released on November 16, 2026. It has 224B total parameters with 12B active per token, accepts text and images, and offers a 1M-token context window. You call it as hy4-flash on Tencent Cloud TokenHub for $0.07 per million input tokens, at half price until November 30.

Contents

What Hy4 flash is

Tencent’s launch post calls Hy4 flash its most cost-effective Hy model to date. It has 224B total parameters, including 64B Engram parameters, and activates 12B parameters per token. Like Hy4, it is natively multimodal and supports a 1M-token context window.

Tencent describes four areas of cost work behind the model:

  • Architecture: hybrid attention, LatentMoE and Engram, which Tencent says cut KV cache size to 1/24 of Hy3 and 1/6 of Hy4.
  • Post-training: self-distillation from Hy4, plus reinforcement learning that Tencent says halves average end-to-end output tokens.
  • Inference: megakernels, 4-bit quantization and DFly, which Tencent says push decoding past 150 tokens per second.
  • Evaluation: WorkIndex, an internal benchmark based on typical WorkBuddy tasks such as slide creation, document editing and spreadsheet repair.

Tencent reports that Hy4 flash slightly outperforms Hy4 preview overall, and that its average cost per WorkIndex task is 1/13 of Hy4 preview, 1/20 of Hy4 and 1/8 of DeepSeek-V4.1-Flash-1017 at peak pricing. In a blind test with 85 internal WorkBuddy users across 259 tasks, Tencent says user satisfaction was close to Claude Haiku 5.5. These are Tencent’s own results, not independent tests.

Hy4 flash vs. Hy4 and other flash models

The TokenHub model pricing page (Singapore site) lists the following list prices per million tokens:

Model (API ID) Input Output Cache hit
Hy4 flash (hy4-flash) $0.07 $0.281 $0.003
Hy4 (hy4) $0.834 $2.501 $0.042
Hy3 (hy3) $0.132 $0.528 $0.033
GLM-5.3-Flash (glm-5.3-flash) $0.15 $0.50 $0.03
DeepSeek-V4.1-Flash-1017, vendor direct (off-peak / peak) $0.15 / $0.30 $0.60 / $1.20 $0.003 / $0.006

The same page says Hy4 flash usage is billed at 50% of list price until 12:00 on November 30, 2026 (Beijing Time), so the current rates are $0.035 input and $0.1405 output, as the Hy4 flash model page also shows. Hy4 has its own 50% discount through December 31, 2026. For background on the larger model, see our Tencent Hy4 guide; for competing low-cost models, see our DeepSeek V4.1 Flash guide and MiniMax-M3.1-Flash guide.

Context window and rate limits

The TokenHub model page and the Hy API guide list these limits for hy4-flash:

  • Context window: 1M tokens (shown as 1024K on the model page).
  • Maximum input: 960k tokens.
  • Maximum output: 64k tokens.
  • Maximum rate: 300 requests per minute and 10,000,000 tokens per minute.
  • Input: text and images. Output: text only.

These input and output caps are the same as Hy4. The older Hy3 has a 256k context window and 128k maximum output.

How to make your first call

TokenHub uses the same base URL for every model and supports three protocols: OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. The steps below follow the TokenHub quick start and the language model API overview:

  1. Sign up for a Tencent Cloud account, complete identity verification and activate TokenHub in the console.
  2. Enable pay-as-you-go billing for Hy4 flash in the Online Inference list.
  3. Create an API key on the API Key page. You can grant it access to all models or a defined set.
  4. Set the base URL for your region: https://tokenhub-intl.tencentcloudmaas.com/v1 (Singapore), https://tokenhub.tencentcloudmaas.com/v1 (Guangzhou) or https://tokenhub-us.tencentcloudmaas.tech/v1 (Silicon Valley).
  5. Send a request to /chat/completions, /responses or /messages with model set to hy4-flash. Use Authorization: Bearer for OpenAI paths and x-api-key for the Anthropic path.

Keys are tied to a site. The overview notes that Guangzhou and Singapore are independent, so a key from one returns error 401002 on the other. Because TokenHub accepts the Anthropic protocol, tools that use either protocol can point to the same base URL.

Thinking, tools and image input

The model page lists reasoning, function calling, structured outputs and prompt caching. The Hy API guide adds these details:

  • Thinking: enable it with "thinking": {"type": "enabled"}. The reasoning appears in reasoning_content. The default reasoning_effort for hy4-flash is high; in the OpenAI Python SDK, pass it through extra_body.
  • Tool calls: when thinking is on, send back the previous reasoning_content with each tool result to keep the chain of thought intact.
  • Structured output: use response_format with a JSON Schema.
  • Images: PNG, JPEG, WEBP or non-animated GIF, up to 10 MB each and up to 600 images per request. The recommended resolution is 128×128 to 2048×2048, and images over 5500×5500 are rejected. A detail setting of low scales images to 512×512 to save tokens.

Outside the API, Tencent says Hy4 flash is free in WorkBuddy until November 30. For Tencent’s open-source coding-agent platform, see our Tencent LoopForge guide. For a larger new Chinese flagship, compare our Qwen4-Max guide.

Limitations

  • No open weights yet. Tencent plans an open-weight release but gives no date.
  • Not Tencent’s pick for heavy coding. Tencent recommends the full Hy4 for long-horizon software engineering.
  • Discount ends soon. Budget for list prices after November 30, 2026.
  • Vendor benchmarks. WorkIndex and the user tests are Tencent’s internal evaluations.
  • OpenRouter pending. Tencent says the listing is coming soon.

Frequently asked questions

When was Hy4 flash released?

Tencent announced Hy4 flash on November 16, 2026. It is available the same day on Tencent Cloud TokenHub and in Tencent apps including WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis. Tencent says an OpenRouter listing is coming soon.

How much does the Hy4 flash API cost?

The TokenHub list price is $0.07 per million input tokens, $0.281 per million output tokens and $0.003 per million cached input tokens. Until 12:00 on November 30, 2026 (Beijing Time), TokenHub bills Hy4 flash at 50% of list price.

What is the Hy4 flash context window?

TokenHub lists a 1M-token context window for hy4-flash, with up to 960k input tokens and up to 64k output tokens per request. The same model page lists 300 requests and 10 million tokens per minute as maximum rate limits.

Is Hy4 flash open source?

Not yet. Tencent says it plans to release Hy4 flash as an open-weight model, but it has not given a date. For now you can use it through TokenHub or Tencent's apps.

Should I use Hy4 flash or Hy4?

Tencent recommends Hy4 flash for most everyday tasks such as deep search, document processing, slide creation, simple frontend demos and computer use. It still recommends the full Hy4 for complex work such as long-horizon software engineering.

Sources

Related answers

Leave a Comment