Tag: TokenHub

  • Tencent Launches Hy4 flash, a Lower-Cost Everyday Model (2026)

    Tencent Launches Hy4 flash, a Lower-Cost Everyday Model (2026)

    Takeaway: On November 16, 2026, Tencent introduced Hy4 flash, a lower-cost, natively multimodal everyday model in the Hy4 family. The API ID is hy4-flash. Tencent Cloud’s international TokenHub pricing page, last updated the same day, lists $0.07 per million input tokens, $0.281 per million output tokens, and $0.003 per million cached-input tokens. Official model docs give a 1-million-token context window, with a 960K-token maximum input and a 64K-token maximum output. Through November 30, Tencent says WorkBuddy access is free, and TokenHub is settling Hy4 flash inference at 50% of the list price until 12:00:00 Beijing Time that day. This is a cost-tier launch next to the flagship Hy4 model, not a replacement for it, and it is distinct from this week’s Qwen4-Max and Gemini 4 Hydrogen posts.

    Tencent’s launch post calls Hy4 flash its most cost-effective Hy model to date. It describes a 224-billion-parameter system, including 64 billion Engram parameters, that activates 12 billion parameters per token, and says the model is natively multimodal with the same 1-million-token context window as Hy4. The useful comparison for API buyers is the official price table, not a headline about “cheap AI.” List prices for flagship Hy4 on the same TokenHub page are $0.834 / $2.501 / $0.042 per million input, output, and cached-input tokens.

    Where Tencent says to use it

    The launch post recommends Hy4 flash for most everyday productivity work: deep search, document processing, slide creation, simple frontend demos, and computer use. It continues to recommend flagship Hy4 for complex workloads such as long-horizon software engineering, where Tencent says the larger model still has a higher intelligence ceiling. Tencent’s Hy API guide, also updated November 16, 2026, describes hy4-flash in similar terms: “a leaner, more efficient everyday workhorse, ideal for high-volume, large-scale tasks such as daily office work.”

    That split matters if you are choosing a default model. Hy4 flash is not positioned as the new flagship. It is the lower-cost everyday option in the same family, with the same documented context window and the same high-level capability list on TokenHub: deep reasoning with preserved thinking, structured output, function calling, caching, and image understanding.

    Official price, discount, and limits

    Use the TokenHub list prices as the starting point, then apply the time-limited discount only if your usage falls inside the official window. The international pricing page, last updated November 16, 2026, at 12:15:10, lists Hy4 flash at $0.07 per million input tokens, $0.281 per million output tokens, and $0.003 per million cached-input tokens. Those are USD list prices. Tencent’s China TokenHub materials list the same model at ¥0.5 / ¥2 / ¥0.02 per million input, output, and cache-hit tokens. Do not mix the two currencies in a single cost estimate.

    The same international pricing page carries an explicit limited-time note: from the date of the notice until 12:00:00 on November 30, 2026, Beijing Time, Hy4 flash inference fees are settled at 50% of the list price. The discount “applies only to usage actually incurred during the above period.” After that cutoff, fees return to the list price. That is narrower than a vague “50% off in November.” If you are planning a migration or a benchmark run, the relevant clock is Beijing Time, and only tokens consumed before noon on November 30 are covered.

    The launch post also says users can access Hy4 flash for free in WorkBuddy through November 30. That is a product-side promotion, not the same rule as the TokenHub API discount. If you are a WorkBuddy user, check the product itself for how the free period is applied. If you are calling the API, the TokenHub pricing note is the rule that matters.

    Tencent Cloud’s official model list, updated the same day, gives these hard limits for hy4-flash:

    • Context window: 1 million tokens
    • Maximum input: 960K tokens
    • Maximum output: 64K tokens
    • Supported capabilities: deep reasoning (preserved thinking), structured output, function calling, caching, and image understanding

    The 1-million-token context window is the same figure Tencent lists for flagship Hy4. The input and output caps are also the same: 960K and 64K. A long document can therefore fit in either model’s window, but the output budget is still 64K tokens. If your workload is long-running generation rather than long-context reading, that output cap is the constraint to test first.

    How you call it

    The Hy API guide says hy4-flash and hy4 are compatible with the OpenAI Chat Completions API, OpenAI Responses, and Anthropic Messages API protocols. The international example endpoint in that guide is https://tokenhub-intl.tencentcloudmaas.com/v1. Image-understanding examples in the same document also show a tokenhub.tencentmaas.com host; use the endpoint for the region and console in which you created the API key, rather than copying a host from a screenshot.

    Tencent says Hy4 flash and Hy4 can take images directly in conversation, for work such as image description, OCR, chart interpretation, and screenshot analysis. The guide lists PNG, JPEG, WEBP, and non-animated GIF, a 10 MB per-image cap, and up to 600 images per request, including multi-turn history. It recommends low detail when you only need a rough reading and high or original when you need small text, tables, or screenshot detail. Images larger than 5500×5500 are rejected; the model scales inputs toward about 2048×2048, so uploading a larger file does not, according to Tencent, reveal more detail.

    Deep reasoning defaults to high for hy4-flash, as it does for hy4, hy4-preview, and hy3. The guide describes preserved thinking and interleaved tool-calling, including the need to send reasoning_content back on later turns. Those are integration details, not a promise that thinking is always on in every product surface. If you are wiring this into an existing OpenAI- or Anthropic-compatible client, read the current guide for the protocol you actually use.

    Where it is available, and what is still planned

    The launch post says Hy4 flash is available across Tencent Cloud TokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima, Marvis, “and more,” and that it will launch on OpenRouter soon. Tencent also says it plans to release Hy4 flash as an open-weight model and to ship frequent updates. Those last two items are plans, not a download page or a date. As of this writing on November 16, 2026, the official materials we reviewed do not give a public open-weight release date or confirm that OpenRouter access is already live.

    What Tencent claims on quality — and how to read it

    The official launch post includes Tencent’s own product-side comparisons. In Yuanbao workflows, it says Hy4 flash boosts agent performance by 13.6% over Hy3 and records a 7.56% gain in human side-by-side (GSB) agent evaluations, standing out for more thorough task completion, clearer presentation, and more credible, professional outputs. On general chat tasks, Tencent says it outperforms GLM-5.3-Flash by 2.52%, with the most notable gains in multi-turn dialogue. Those figures are Tencent’s. They are not independent third-party leaderboard results, and the launch post does not present a full public evaluation card with methodology, prompts, or raw scores.

    Parameter counts in the same post — 224 billion total, including 64 billion Engram parameters, and 12 billion active per token — are also Tencent’s description of the model. They help explain why the company is pricing Hy4 flash far below Hy4, but they are not a substitute for measuring your own latency, quality, and token use. If a later official model card or open-weight release adds architecture or evaluation detail, that would belong in an update. Until then, treat quality claims as company-reported and price/limit figures as the part of the launch that is documented in TokenHub.

    How this compares with other everyday models this week

    Hy4 flash lands on the same day as Alibaba’s Qwen4-Max general-availability launch, and in the same week as Google’s Gemini 4 Hydrogen. The useful distinction is not “another flash model,” but the official price and the intended job. Qwen4-Max is a flagship QwenCloud model with Singapore list prices of $1.50 / $4.50 per million tokens. Gemini 4 Hydrogen is Google’s fast everyday Gemini 4 model, with a standard paid price of $1.50 / $7.50 per million tokens. Hy4 flash’s TokenHub list price is an order of magnitude lower than both, at $0.07 / $0.281, before the temporary 50% API discount. That does not make it a drop-in substitute. The three models sit in different APIs, with different context caps, feature sets, and quality claims, and Tencent itself still points heavier software-engineering work at flagship Hy4.

    Hy4 flash facts at a glance

    ItemWhat official sources say
    AnnouncementNovember 16, 2026, Tencent Hy launch post
    Model IDhy4-flash
    List price (international TokenHub, USD per million tokens)Input $0.07; output $0.281; cached input $0.003
    List price (China TokenHub, CNY per million tokens)Input ¥0.5; output ¥2; cache hit ¥0.02
    Limited-time API discount50% of list price until 12:00:00 November 30, 2026, Beijing Time; only usage incurred during the window
    WorkBuddy promotionFree access through November 30, according to the launch post
    Context / max input / max output1M / 960K / 64K tokens
    Documented capabilitiesDeep reasoning (preserved thinking), structured output, function calling, caching, image understanding
    Tencent’s size description224B total parameters, including 64B Engram parameters; 12B active per token
    Where Tencent says it is liveTokenHub, WorkBuddy, Yuanbao, WeCom, QQ, ima, Marvis, and more; OpenRouter “soon”
    Still planned, not datedOpen-weight release; frequent updates
    Quality comparisonsTencent-reported Yuanbao and general-chat figures; not independent benchmarks

    FAQ

    What is Tencent Hy4 flash?

    Hy4 flash is Tencent’s lower-cost everyday model in the Hy4 family, announced November 16, 2026. The API ID is hy4-flash. Tencent describes it as natively multimodal, with a 1-million-token context window, and recommends it for everyday productivity rather than as a replacement for flagship Hy4.

    How much does Hy4 flash cost?

    Tencent Cloud’s international TokenHub pricing page lists $0.07 per million input tokens, $0.281 per million output tokens, and $0.003 per million cached-input tokens. China TokenHub materials list ¥0.5 / ¥2 / ¥0.02 for the same three items. Through 12:00:00 Beijing Time on November 30, 2026, TokenHub is settling Hy4 flash inference at 50% of the list price for usage incurred during that window.

    Is Hy4 flash free in WorkBuddy?

    Tencent’s launch post says users can access it for free in WorkBuddy through November 30. That is separate from the TokenHub API discount, which ends at noon Beijing Time on the same date and applies only to usage incurred during the promotional period.

    What is the context window?

    Official TokenHub model docs list a 1-million-token context window, a 960K-token maximum input, and a 64K-token maximum output. Those caps match the official figures for flagship Hy4.

    Is this a replacement for Hy4?

    No. Tencent recommends Hy4 flash for most everyday productivity tasks and still recommends Hy4 for complex workloads such as long-horizon software engineering. The two models share a documented 1M context window, but Tencent says Hy4 has a higher intelligence ceiling.

    Which APIs does it support?

    The official Hy API guide says hy4-flash is compatible with the OpenAI Chat Completions API, OpenAI Responses, and Anthropic Messages API protocols. It also documents image understanding, preserved thinking, function calling, and structured output. Use the TokenHub endpoint for the region where your API key was created.

    Are the benchmark numbers independent?

    No. The Yuanbao agent figures and the general-chat comparison with GLM-5.3-Flash in the launch post are Tencent’s. This article does not treat them as independent third-party results. The prices, context limits, and discount window come from official TokenHub documentation.

    Can I download the weights, or use it on OpenRouter today?

    Not according to the official materials reviewed for this article. Tencent says it plans to release Hy4 flash as an open-weight model and that the model will launch on OpenRouter soon. Those are plans, without a public date in the launch post or TokenHub docs we reviewed on November 16, 2026.