By Jonathan West · Updated Nov 17, 2026

Tencent Hy4 Flash Explained

Tencent's everyday-work model, with separate USD and CNY price listings and a trial verdict as of November 17, 2026.

By Jonathan West · Updated Nov 17, 2026

Hy4 flash is Tencent's low-cost multimodal model for everyday work, listed at US dollars (USD) 0.07 per million input tokens and USD 0.281 per million output tokens on Tencent Cloud TokenHub. Tencent released it on November 16, 2026. It is worth a trial for document, search and slide tasks. Tencent recommends Hy4 instead for long software-engineering work.

Tencent's launch announcement is in Chinese, while the TokenHub page lists prices in US dollars. Tencent's Hy site also lists prices in Chinese yuan, so the two listings need their currencies kept separate. The speed, task-cost and user-satisfaction figures below are Tencent's reported results. They do not establish how Hy4 flash will perform on a business's own documents.


What Tencent Hy4 Flash Is and What It Costs

Hy4 flash is Tencent's multimodal Hy model for everyday work, with 224 billion total parameters, 12 billion active parameters and a one-million-token context length. Tencent says 64 billion of the total are Engram parameters. The model's application programming interface (API) identifier is hy4-flash. Its listed input and output rates make it a candidate for document, search and slide workflows, but the cost of a completed task depends on the tokens that task uses.

As listed on the TokenHub price page, Hy4 flash costs USD 0.07 per million input tokens, USD 0.281 per million output tokens and USD 0.003 per million cached tokens. The page has Singapore, Guangzhou and Silicon Valley region tabs. Those tabs have not been verified individually for this listing. Use the figures as the TokenHub page lists them, rather than assuming the same rate applies in every region.

Tencent's Hy4 flash announcement gives a separate Chinese yuan (CNY) listing per million tokens: CNY 0.02 for cached input, CNY 0.5 for input and CNY 2 for output. Those are prices from Tencent's Hy site, while the USD figures come from TokenHub.

Tencent calls Hy4 flash its best-value Hy model so far and recommends it for most everyday tasks. That recommendation is the basis for a limited trial, rather than a reason to route every task to it. Tencent points businesses doing long-horizon software engineering toward Hy4, which it describes as having a higher intelligence ceiling.

  • TokenHub listing: USD 0.07 input, USD 0.281 output and USD 0.003 cache per million tokens
  • Tencent Hy-site listing: CNY 0.5 input, CNY 2 output and CNY 0.02 cached input per million tokens
  • Model size: 224 billion total parameters, including 64 billion Engram parameters, with 12 billion active parameters
  • Context length: one million tokens, according to Tencent

Hy4 Flash Specs, Channels and Availability

Hy4 flash is live in Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis, according to Tencent's November 16 announcement. Tencent says OpenRouter access is coming soon and that it plans to open-source the model. Neither statement means that access or open weights have shipped. If either is a requirement for a deployment, wait for Tencent to ship that access.

ItemTencent-published detailDecision relevance
API identifierhy4-flashSelects Hy4 flash for an API request
Model size224B total (64B Engram), 12B activeDescribes Tencent's model configuration
ModalitiesMultimodalSupports Tencent's listed visual-understanding work
Context length1M tokensStates Tencent's published context length
Live channelsTokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and MarvisShows where Tencent says Hy4 flash can be used
OpenRouterComing soonDo not plan a current rollout around it
Open-source releasePlannedNo released weights are established by the announcement
WorkBuddy offerFree for a limited time, until November 30Offers a trial channel within Tencent's stated window

Tencent's TokenHub notice gives the discount cutoff precisely: "Hy4 flash Limited-Time Discount Notes: From the date of this notice until 12:00:00 on November 30, 2026 (Beijing Time), Hy4 flash model inference fees are settled at 50% of the list price. The discount applies only to usage actually incurred during the above period. After the period expires, the model fees will be settled at the list price again." The TokenHub page does not establish a separate post-discount USD rate for this article to quote. Budget against the listed rate and the notice's timing without inventing another USD price.

Tencent describes hybrid attention, LatentMoE and Engram as parts of Hy4 flash's architecture. It also reports a key-value (KV) cache one-sixth the size of Hy4's. Tencent recommends Hy4 flash for everyday work. Its reported decode speed is not a completion time for a particular document task. For cross-vendor listed rates, use the AI model price list alongside the TokenHub listing.

  • WorkBuddy: Tencent says it is free for a limited time until November 30, 2026
  • TokenHub discount: applies to usage incurred before 12:00:00 on November 30, 2026, Beijing time
  • OpenRouter and open-source access: announced as future plans

What Tencent Reports About Speed and Cost

Tencent reports that Hy4 flash decodes at more than 150 tokens per second. Tencent attributes its inference performance in part to a megakernel, 4-bit quantization and DFly optimizations. Decode speed describes token generation. It is not a measured completion time for every document or agent task. A trial should time the full task, including the work before and after text generation.

Tencent evaluates Hy4 flash on WorkIndex, an internal collection built from typical WorkBuddy tasks and combined with human evaluation. Its eight sets cover slide creation, slide revision, slide co-writing, Markdown co-writing, document editing, spreadsheet repair, audio and video processing, and visual understanding. Tencent says WorkIndex focuses on output quality, single-turn output length and end-to-end time. A WorkIndex cost ratio therefore describes Tencent's evaluation, rather than a guaranteed reduction on a business's TokenHub bill.

On WorkIndex, Tencent reports a cost per task for Hy4 flash of one-thirteenth of Hy4 preview's, one-twentieth of Hy4's and one-eighth of DeepSeek-V4.1-Flash-1017's. Tencent specifies the max tier and busy-hour price for the DeepSeek comparison. DeepSeek's release notes, summarized on the DeepSeek V4.1 Flash explained page, describe DeepSeek-V4.1-Flash-1017 as the October 17, 2026 update that the deepseek-flash API name now serves. Tencent also reports that Hy4 flash scored slightly better overall than Hy4 preview in its evaluation.

Tencent reports a separate blind test involving 85 internal WorkBuddy users, 27 task categories and 259 everyday tasks. Tencent says Hy4 flash satisfaction was close to Claude Haiku 5.5, its cost per task was lower than the other models in that test, and its task time was the shortest. These are Tencent-run results. The announcement lists task types to examine, but a buyer still needs to compare answers and elapsed time on the work they intend to route.

Tencent's quoted WorkBuddy product manager reports average task time 70 percent lower than Hy4's and average cost about one-fifteenth of Hy4's for that product team's work. The manager also reports a small drop on long-horizon coding tasks. That qualification supports Tencent's recommendation to reserve Hy4 for complex software-engineering work.

  • Decode speed: above 150 tokens per second, reported by Tencent
  • WorkIndex cost per task: one-thirteenth of Hy4 preview, one-twentieth of Hy4 and one-eighth of DeepSeek-V4.1-Flash-1017 under Tencent's stated comparison conditions
  • Blind test: 85 internal users completed 259 everyday tasks across 27 categories, reported by Tencent

Hy4 Flash Workflows for Business

Tencent recommends Hy4 flash for deep search, document processing, slide generation, simple front-end display and computer use. A business can turn those categories into a trial by choosing one existing document task and defining the required output before running it. For example, choose a document that needs a written result, then compare the result, elapsed time and token usage with another model under consideration. The trial answers a narrower question than Tencent's benchmarks: whether Hy4 flash handles that document acceptably.

Search work needs its own check. Give Hy4 flash the same search task and the same standard for an acceptable answer that you give the comparison model. For slide work, decide whether the job is creating a deck, revising one or co-writing one. Tencent separates those jobs in WorkIndex. Compare them as separate tasks.

Computer use and a simple front-end display deserve separate trials from document processing. Tencent lists both as everyday-work use cases, but a satisfactory written answer does not establish that the model completes an interface task. Set a finish condition for the chosen task and review the completed work. If the task instead requires long software-engineering work, start the comparison with Hy4. Tencent recommends Hy4 for that category.

At Layer3Labs, we build and operate AI systems and workflow automation inside small and mid-sized business teams. For a Hy4 flash trial, compare two outputs from the same task against the same finish condition. Tencent's published rates can then be considered alongside the task result without treating its WorkIndex ratios as the team's own expected savings.

  • Document task: specify the required written output and review whether it is complete
  • Slide task: distinguish creation from revision or co-writing
  • Interface task: define a finish condition before evaluating computer use or front-end display
  • Long software-engineering task: include Hy4 in the comparison, as Tencent recommends

Who Should Not Use Hy4 Flash and What Would Change the Answer

Hy4 flash is a poor first choice for a team whose main job is long-horizon software engineering. Tencent recommends Hy4 for that work and says Hy4 has a higher intelligence ceiling. Tencent's quoted WorkBuddy product manager also reports a small drop in Hy4 flash results on long-horizon coding tasks. Start with Hy4 for that workload, then include Hy4 flash only if a representative task shows that its output meets the same finish condition.

Teams that require independent benchmark evidence should wait before treating Tencent's WorkIndex ratios or blind-test satisfaction as a procurement finding. Tencent built WorkIndex from its own WorkBuddy tasks, and the blind test used internal WorkBuddy users. Those reports support a trial. They cannot establish the result for a different team's documents. Independent results that contradict Tencent's reports would weaken the case for Hy4 flash even on the everyday tasks Tencent recommends.

Access requirements can change the answer too. Tencent says OpenRouter support is coming soon and an open-source release is planned. A team that requires either route should use an available model until Tencent ships the required access. If those routes launch, Hy4 flash becomes an option for teams whose deployment rules exclude its currently listed channels.

The price decision has a date attached. TokenHub says its 50 percent settlement discount applies only to qualifying Hy4 flash usage before 12:00:00 on November 30, 2026, Beijing time, after which fees return to list price. A changed TokenHub rate or the end of that offer would change a cost comparison. For the current decision, Hy4 flash merits a bounded everyday-work trial on Tencent's reported evidence, while Hy4 remains Tencent's recommendation for long software-engineering tasks. Try Hy4 flash on one document workflow through WorkBuddy before November 30, then compare its output and task time with a model you are considering.

  • Choose Hy4 first when long-horizon software engineering is the main task
  • Wait for independent results if Tencent-run evaluations cannot support the decision
  • Wait for a release if OpenRouter access or open weights are required
  • Revisit cost assumptions when the TokenHub discount ends or its listed rates change

Frequently Asked Questions

  • Hy4 flash is Tencent's multimodal Hy model for everyday work, released November 16, 2026. Tencent lists 224 billion total parameters, 12 billion active parameters and a one-million-token context length. It recommends Hy4 flash for tasks including document processing and slide generation.
  • As listed on the Tencent Cloud TokenHub price page, Hy4 flash costs USD 0.07 per million input tokens, USD 0.281 per million output tokens and USD 0.003 per million cached tokens. Tencent's Hy site separately lists CNY 0.5 for input, CNY 2 for output and CNY 0.02 for cached input per million tokens. TokenHub also posts a time-limited settlement discount. Neither listing supplies a separate discounted USD figure for this answer.
  • Tencent says WorkBuddy is free for a limited time until November 30, 2026. Tencent lists paid Hy4 flash API token rates on its Hy site and TokenHub. TokenHub's 50 percent settlement discount applies to eligible usage before 12:00:00 on November 30, 2026, Beijing time.
  • Tencent says Hy4 flash is live on Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima and Marvis. Tencent describes OpenRouter access as coming soon, so it is not a currently listed live channel.
  • Tencent plans to open-source Hy4 flash, but its November 16 announcement describes the release as a plan. Teams that require open weights should wait for Tencent to publish them and their terms.

Considering Hy4 Flash for a Business Workflow?

Book a free 30-minute AI workflow audit with Layer3Labs. We can help scope a document-task trial and the criteria for comparing models.

Book a Consultation