Tencent
Hy4 flash
a model for lower-cost everyday workflows

Key facts
- 16 Nov 2026Tencent announcement
- Released
- Up to 1Mtokens
- Context
- 224Bincluding 64B Engram
- Total parameters
- 12Bper token
- Active parameters
Hy4 flash is Tencent’s multimodal AI model for everyday work. It supports up to one million context tokens and uses 12 billion active parameters from a larger model.
Hy4 flash is Tencent’s AI model for everyday tasks that need multimodal understanding and lower-cost responses. Tencent announced it on 16 November 2026; CloudBase added it on 17 November. It is a separate release from Hy4 and Hy4 preview.
A smaller part of the model runs for each token
Tencent reports 224 billion total parameters, including 64 billion Engram parameters, with 12 billion active parameters and a context window of up to one million tokens. Parameters are learned model values; the active count describes the subset used during a step of generation.
The model accepts multimodal input. Tencent positions it for routine work across its products, including WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis. TokenHub offers developer access; product allowances follow their own terms.
Launch offers have their own deadlines
Tencent’s release quotes approximate dollar equivalents of $0.070 input, $0.281 output and $0.003 cached input, per million tokens. These are the provider’s quoted launch figures; confirm the applicable region, currency and promotion in the billing service before estimating costs.
Tencent advertises a TokenHub 50% promotion and a WorkBuddy offer through 30 November. Separately, Vercel’s AI Gateway announcement offers its own 50% promotion through 29 November. The services and deadlines are distinct.
Availability and evaluation claims need their own labels
Tencent lists OpenRouter access as forthcoming and says weights are planned. Those plans should be distinguished from the hosted access already announced. Its internal WorkIndex comparisons are vendor evaluations; they do not establish performance on every workload.
For a practical comparison, test representative inputs, tool calls, response time and billing through the service you intend to use. See the LLM directory for other releases and the Hy4 guide for the production family.
More in Large Language Models
All LLMs →- Google DeepMindGemini 4 Argonlonger reasoning, coding and cyber defence
- OpenAIGPT-5.6the flagship from 9 July to 3 September 2026
- OpenAIGPT-6 Astrathe new generation, rolling out from 3 September 2026
- AnthropicClaude Fable 5.2An AI model for difficult coding, research and long-running tasks
- Google DeepMindGemini 4 CarbonCoding, research and tasks with many steps, now available more widely
- OpenAIGPT-6.1 Solthe upgraded mid-price GPT-6 model, released 29 September 2026