Model release · Tencent
Tencent releases Hy4 flash with open weights: free for a limited time on OpenCode and WorkBuddy, 50% off on the TokenHub API, and now on OpenRouter
- Updated
- Launched
- Ends
Tencent says Hy4 flash is its cheapest model for everyday productivity. In human blind tests, it says the model matches Claude Haiku 5.5 (high) on daily work while cutting task cost 18 times. API pricing per million tokens is $0.003 for cached input, $0.070 for input and $0.281 for output. OpenCode, including OpenCode Go, is offering it free for about two weeks with near-unlimited usage, and it does not count against Go quota; in OpenCode, pick Hy4 flash Free (hy4-flash-free) with /models. OpenCode also cites a 1M context, multimodal input and zero data retention. Tencent Hunyuan says WorkBuddy is offering it free for a limited time. A WorkBuddy-authorized note says both the Chinese and international apps are included, with a daily free quota rather than unlimited use. After the quota runs out, users can continue by spending credits, and may have to wait in a queue when the service is busy. Image and video generation is switched to other models and still costs credits as usual. The same note says the earlier free period for Hy4 (not Hy4 flash) in WorkBuddy has ended, and that Hy3 will be removed from the app. Tencent Cloud TokenHub has the API at 50% off (model ID hy4-flash). List prices are 0.5 yuan per million input tokens, 2 yuan output and 0.02 yuan cached input, discounted to 0.25, 1 and 0.01 yuan. No coupon is needed; the discounted rate applies when you call the API. Hy4 flash is also now on OpenRouter (tencent/hy4-flash), with a 1M-token context window, image input and optional reasoning. Tencent has since released the weights: instruct, FP8-quantized and pre-trained base versions are on Hugging Face, ModelScope and GitCode, with 220B total parameters (including 64B Engram) and 12B active, and they can be served with vLLM, SGLang or Transformers. The Tencent Hy License 1.0 is derived from Apache 2.0: a MaaS business with over $50 million in worldwide annual revenue, or using it in a product with over 100 million monthly active users, needs a separate written license from Tencent; use solely for internal purposes, with nothing made available to third parties, is exempt.
Availability
Available through the Tencent Hy API. OpenCode, including OpenCode Go, is offering it free for about two weeks with near-unlimited usage, and it does not count against Go quota; in OpenCode, pick Hy4 flash Free (model ID hy4-flash-free) with /models. WorkBuddy's Chinese and international apps offer it free for a limited time with a daily quota; after that, further use costs credits, and users may wait in a queue when it's busy. Image and video generation is switched to other models and still costs credits as usual. Tencent Cloud TokenHub has the API at 50% off (model ID hy4-flash): list prices are 0.5 yuan per million input tokens, 2 yuan output and 0.02 yuan cached input, discounted to 0.25, 1 and 0.01 yuan. No coupon is needed; the discounted rate applies when you call the API. Also available on OpenRouter (tencent/hy4-flash), with a 1M-token context window, image input and optional reasoning. Weights are open on Hugging Face, ModelScope and GitCode in instruct, FP8-quantized and pre-trained base versions under the Tencent Hy License 1.0: a MaaS business with over $50 million in worldwide annual revenue, or using it in a product with over 100 million monthly active users, needs a separate written license from Tencent; purely internal use is exempt.