FOUNDING OFFERSign up to unlock 35% off your first plan.

What Is Hy4 flash? Multimodal Specs, WorkIndex Tasks, and Pricing

SUMMARY

Hy4 flash is Tencent's November 16, 2026 multimodal AI model built for fast, low-cost office and media workflows rather than heavy software engineering. For business teams, the release highlights a practical divide between automated text or spreadsheet drafts and customer-ready marketing assets: rapid model output still lacks calibrated product proportions, consistent lighting, and verified offer wording before publication. Independent merchants do not need to compare model benchmarks, write prompts, or assemble edits themselves. Instead, shops can upload existing phone photos to Stillmore, where the creative team may evaluate tools such as Hy4 flash alongside photography direction and post-production to deliver checked promotional images, short videos, and captions for the merchant to publish.

Lan He

Stillmore Blog cover for Hy4 flash, featuring glass triangle and cyan vertices.
Contents

What Hy4 flash Is and How Its Architecture Works

Hy4 flash is a sparse multimodal model designed to process large files and mixed media inputs while activating only a fraction of its stored weights per generated token. Separating total parameter capacity from per-token compute helps keep routine workplace sessions responsive without running full-model calculations on every word.

Per Tencent's announcement, Hy4 flash pairs 224B total parameters (64B Engram, 12B active) with hybrid attention, LatentMoE, 1M-token context, and >150 tps decoding. API pricing runs US$0.070 input, US$0.281 output, and US$0.003 cached per million tokens; WorkBuddy access is free (TokenHub half-off) through November 30, with Hy4 kept for complex coding and open-sourcing planned.

In practice, consider an independent merchant reviewing a multi-page supplier catalog alongside several product reference images in one thread. In general sparse routing architectures, activating a smaller share of parameters per token reduces the compute required at each decoding step compared with running every weight in a dense model of similar total capacity. Even when a model processes mixed text and visuals quickly, its output remains an unverified working draft rather than a checked commercial asset.

Evaluation Factor What It Means for Operations Acceptance Question for Your Team
Per-token compute vs. capacity Balances broad general recall with lower serving overhead Does the assistant handle multi-file drafts fast enough for daily staff use?
Input modality mix Accepts text, documents, and reference images in one session Can staff drop a price sheet and product photo into one thread cleanly?
Context capacity Holds long catalogs or handbooks without splitting files Does the model track earlier SKU rows accurately by the end of a long sheet?
Draft vs. finished quality Speeds up outlines and cleanup, not final brand layouts Who verifies numbers, label spelling, and product proportions before release?

Why Hy4 flash Matters for Everyday Workflows and API Costs

Lightweight multimodal releases matter to cost-sensitive operators because routine internal requests, such as summarizing vendor emails, repairing spreadsheet formulas, and drafting internal outlines, accumulate high daily token volume where heavier reasoning models quickly strain budgets.

Looking at standard API billing structures, where fresh input, repeated cached context, and newly generated output tokens carry distinct per-million rates, concise responses directly lower recurring spend. In Tencent's release notes, reinforcement learning is credited with halving average output tokens, complementing lower-priced cached reads.

Imagine a neighborhood bakery manager sending the same weekly allergen guide and catering price list to an internal assistant every morning to draft shift notes. When cached context keeps repeated store documents inexpensive and concise replies avoid multi-paragraph filler, back-office admin costs stay predictable. However, predictable text billing does not translate into finished storefront marketing, because social videos and posters still demand photography judgment, layout balance, and visual quality checks.

How Hy4 flash Performs Across WorkIndex and Workplace Handoffs

Workplace-oriented model evaluations examine whether an assistant finishes practical office deliverables cleanly and concisely inside everyday software environments, rather than relying solely on abstract academic trivia.

Rather than repeating individual test names, the official release groups performance around internal WorkIndex office tasks, public agent suites, and blind satisfaction testing across Tencent workplace apps.

Consider an independent home-goods shop using a workplace assistant to clean up a messy wholesale spreadsheet and outline a five-slide buyer deck. Even if the assistant formats cells and organizes talking points cleanly, an internal office deliverable is not a customer-facing campaign. Turning those product points into scroll-stopping social media visuals still requires real source photos, visual framing, retouching, and a pre-publish check.

Workflow Stage What a Workplace Assistant Can Draft What Still Requires Creative Production
Product information cleanup Tidies SKU tables, bullet lists, and internal memos Verifying exact consumer offer dates and compliant label wording
Campaign ideation Suggests promotional angles or rough storyboard beats Selecting camera angles and lighting suited to the physical item
Visual asset prep Analyzes uploaded images or describes scene elements Compositing, color matching, and protecting real product shape
Video assembly Drafts short voiceover scripts or text overlays Pacing motion, editing clips in tools like After Effects, and readability checks

Illustrative Business Workflows: Model Output vs. Finished Creatives

A raw model output such as a caption list, slide outline, or unedited generation serves as an internal starting point, not a publishable advertisement for a real storefront.

As documented in Inside Stillmore: Our Creative Production Workflow, turning everyday product photos into finished social creatives depends on coordinating photography direction, Adobe Photoshop compositing, Adobe After Effects motion work, and five pre-delivery checks: product shape, color and surface texture, exact text and offer terms, readable motion timing, and clear focal composition on small screens. Because editing cannot reliably establish packaging details missing from source shots, merchants must also supply clear photos and current offer wording.

Illustratively, picture a small cold-brew coffee brand starting from a casual phone photo of an amber glass bottle on a shop counter. A fast assistant can summarize tasting notes or propose three headline hooks in moments. Producing a polished poster or short-form video, however, means matching the black cap and glass hue to the actual bottle, keeping the printed label sharp during camera movement, placing Sunday's offer deadline clearly on screen, and reviewing the final frame before posting.

Acceptance Check Unchecked Model Output Publish-Ready Brand Deliverable
Reference grounding Generic layout or unverified visual proportions Anchored directly to the merchant's real product or shop photos
Shape and silhouette Edges, caps, or garment cuts may drift Proportions verified against supplied reference angles
Color and surface Lighting shifts may alter packaging tone Material finish and brand colors kept recognizable
Label and offer text Relies on manual user inspection Product names and offer dates checked against merchant copy
Motion readability Static draft or erratic clip timing Paced so key details stay legible on mobile feeds
Small-screen framing Subject can get lost in busy compositions Cropped for clear focus at feed thumbnail size

Hy4 flash vs. Hy4, Hy4 Preview, and Adjacent Models

Choosing between Hy4 flash, heavier models like Hy4, and rival lightweight releases comes down to trading peak reasoning depth for faster turnaround and lower per-task spend.

In Tencent's comparison notes, the team contrasts Hy4 flash against larger internal siblings (Hy4 Preview and Hy4) as well as external flash baselines like DeepSeek-V4.1-Flash-1017 and Claude Haiku 5.5, explicitly directing complex multi-step engineering work toward Hy4.

Imagine an online apparel boutique maintaining both its back-office product spreadsheet and a custom checkout integration. For daily SKU formatting, customer FAQ drafts, and short internal summaries, a compact flash-tier model keeps routine overhead modest without bogging down simple requests. When a developer needs to debug multi-file storefront logic across a complex repository, however, stepping up to a heavier reasoning tier avoids cutting corners on deep state tracking.

Business Need Best-Fit Option Why It Fits Key Limitation to Watch
High-volume memos, spreadsheets, slide outlines Flash-tier multimodal assistant Keeps routine text and document tasks fast and economical Outputs still require human fact-checking and formatting
Long-horizon coding or multi-file software refactors Heavyweight reasoning model Built for deeper multi-step logic and state tracking Higher per-task cost and slower turnaround for simple tasks
Consistent Instagram, TikTok, or Shorts promos Dedicated creative production service Combines real photos, compositing, editing, and visual checks Follows scheduled production cycles rather than instant chat

Where Hy4 flash Fits and Where Its Limits Begin

Fast workplace models fit structured administrative tasks and internal research drafts well, while hitting clear boundaries on deep software engineering, undisclosed licensing terms, and unassisted brand marketing.

Before committing workflows, evaluators should verify whether a release includes a published license and complete pricing terms in text before relying on future distribution plans. Meanwhile, Stillmore's workflow documentation emphasizes that automated tools cannot verify unstated business terms such as weekend promotion deadlines or replace visual consistency checks against real physical goods.

Suppose a local skincare shop expects a single chat assistant to turn rough stock photos into ready-to-run ads overnight. Even accurate visual analysis and quick caption brainstorming cannot guarantee that glass bottle proportions, label typography, and lighting shadows hold together across composited edits. Keeping lightweight assistants on internal prep while protecting pre-delivery creative review prevents distorted packaging from reaching customer feeds.

How to Test Hy4 flash or Offload Content Production

Technical evaluators can test raw model speed directly inside Tencent's promotional platforms, whereas time-pressed business owners who want finished promotional assets can hand off production without touching model settings.

Hands-on developers can explore the model through the live channels and limited-time launch pricing outlined in Tencent's announcement. For independent businesses that would rather serve customers than compare AI tools, write prompts, or edit clips, Stillmore provides a subscription creative agency built around your existing phone photos.

If your goal is benchmarking API latency or internal spreadsheet automation, start with a sample document brief in an evaluation session before committing engineering time. If your real bottleneck is keeping Instagram, TikTok, or YouTube Shorts active between customer orders, upload your product, space, or team photos to Stillmore instead. Stillmore combines photography direction, post-production, and an AI production team that may evaluate models such as Hy4 flash when suited to a brief, delivering checked images, videos, and captions for you to publish when ready.

Conclusion

Hy4 flash fills a specific operational lane for teams prioritizing lightweight, low-overhead internal drafting. That profile suits technical teams and workplace operations testing fast, economical automation. When the priority shifts from back-office drafts to customer-facing social campaigns, reliable results still depend on clear source photos, hands-on editing, and pre-delivery product checks, whether handled by an in-house designer or a dedicated creative team working from photos you already own.

Resources

Source Link Verification Scope
Tencent Corporate News Tencent Launches Hy4 flash Official release date, architecture specs, USD pricing, WorkIndex positioning, and November 30 promo window
Stillmore Production Workflow Inside Stillmore: Our Creative Production Workflow Photography, VFX/post-production, AI team evaluation role, and five pre-delivery checks
Stillmore Service Overview What Is Stillmore? Content Creation for Small Businesses Existing photo inputs, manual vs. automatic creation, and merchant publishing control

Frequently Asked Questions

Do I need to learn or select Hy4 flash to work with Stillmore?

No, you never need to learn, select, or run Hy4 flash or any production software when working with Stillmore. Clients pay for ongoing creative production and finished deliverables rather than self-serve model access. When a brief suits it, Stillmore's AI production team may evaluate and use generative tools alongside photography direction, compositing, and post-production. You simply upload everyday photos and share business details, while the studio handles tool selection, editing, and pre-delivery checks before placing finished assets in your account.

What materials do I provide instead of writing AI prompts?

You supply existing phone photos of your products, shop space, staff, or brand alongside essential business details such as the exact product name and promotion dates. Rather than writing technical prompts yourself, you give the creative team visual references and commercial context to build from. Before uploading any image, zoom in to verify that labels, edges, and key features are clearly visible. Retouching and generative tools cannot reliably establish packaging details hidden by reflections or cropped outside the frame.

How does Stillmore prevent distortions on product labels or packaging?

Stillmore inspects every finished image and video against your supplied reference photos before delivery to catch visual drift early. The pre-delivery check reviews five areas: accurate product shape and proportions, recognizable surface color and texture under new lighting, exact wording matching your supplied product and offer terms, steady motion that keeps details readable, and clear focal composition on small mobile screens. If a cap, seam, or printed label looks off during production, the team refines the piece before sending it.

Hy4 flash targets workplace drafts, so what deliverables does Stillmore actually send me?

Stillmore sends finished promotional images, short videos, and matching post captions directly to My Creations for channels such as Instagram, TikTok, and YouTube Shorts. Productivity assistants help organize internal documents, spreadsheets, and notes, leaving commercial visual design and layout work unfinished. Stillmore coordinates photography direction, Photoshop retouching, After Effects motion work, and pre-delivery checking so you receive complete promotional posts grounded in your supplied photos, without opening design software, video editing timelines, or caption documents yourself.

Can I choose specific promotional ideas myself while leaving routine posts to the team?

Yes, you can switch between manual selection and automatic creation anytime inside My Creations, or combine both within your monthly balance. Browse creative styles in Inspiration to request specific product launches yourself, and enable automatic creation once your Library holds at least 3 photos on Try it or 15 photos on Growth. Both modes share the same Plan Credits, and switching back to manual stops new automatic creations while any work already in progress still finishes.

How does Stillmore Credit pricing compare to paying per million AI tokens?

Stillmore bills through fixed per-deliverable Credits instead of variable token meters, so you know the exact balance required before requesting content. One finished video uses 3,000 Credits and one finished image uses 500 Credits, and you can mix both formats freely within your active balance. Rather than charging separately for fresh inputs, cached context, and repeated output retries, your Credits cover creative direction, retouching, motion editing, caption writing, and pre-delivery checks for each finished promotional piece.

If fast models generate text in seconds, why does Stillmore take up to 5 days?

Raw model speed only generates unverified drafts, whereas Stillmore's 5-day turnaround covers creative direction, Photoshop and After Effects editing, captioning, and pre-delivery review. On Growth automatic mode, your first video arrives within 5 days once you upload at least 15 suitable photos, enable automatic creation, and have payment and Credits ready, continuing every 5 days (late-month uploads may not finish by month-end). Manual requests arrive within 5 days after payment, materials, and submission finish, while Add-on videos ship one at a time within 5 days each.

Do I have to upload fresh product photos every time I want new ads?

No, photos uploaded to your Stillmore Library remain saved for reuse across future promotions without monthly reshoots. A single set of clear product or shop photos can support fresh visual directions over time, whether you submit manual requests from Inspiration or let the team choose assets automatically. You only need to upload new shots when introducing a new item, updating packaging, or supplying a better angle that earlier photos missed, because saved photos stay useful only while they accurately reflect your product.

Does Stillmore post finished videos and images directly to my Instagram or TikTok?

No, Stillmore does not connect to or publish automatically on your social media accounts. When Stillmore describes its service as handling social content, that covers planning, visual selection, creative production, caption writing, and pre-delivery checks. Once your finished images or videos pass review, Stillmore sends an email notification and places the assets in My Creations. You retain full control over your brand channels and choose when, where, and how to publish delivered posts across Instagram, TikTok, or YouTube Shorts.

What happens if my promotion date or offer wording changes mid-production?

Notify Stillmore promptly whenever your product name, offer wording, or promotion end date changes so the team can verify the right text. Reference photos show how a physical item looks, yet they cannot reveal commercial terms such as weekend discounts or closing dates, leaving you as the source for accurate copy. Because pre-delivery checks compare on-screen text against the details you supplied, sending updates early helps avoid mismatches, while formal delivery and revision terms follow the Help Center conditions.

What happens to my unused Credits if I pause or cancel my Stillmore plan?

If you cancel future renewals in Plan and Billing, your subscription remains active until the end of your paid period, after which new monthly Credits stop. Any unused Plan Credits remain available and can accumulate up to a balance cap equal to two months of your plan's allocation, which is 6,000 Credits for Try it or 36,000 Credits for Growth, rather than expiring after two months. Purchased Add-on Credits never expire and remain usable after cancellation, while manual mode stops new automatic work.

About the author

Lan He

I'm Lan. Years of writing promotional and editorial work in the US, mostly for companies with big audiences and bigger expectations — where a single article might reach hundreds of thousands of readers, and where the good ones kept earning traffic long after the campaign ended. I learned what makes a page survive that: one clear audience, one real question, images that show the actual work, and a structure that keeps people reading. Now I write for small business owners who do everything themselves, answering the questions they actually ask: what's worth posting, which photos are worth taking, and how to tell one story across images, video, and words. I've seen good content carry a small company further than its ad budget ever did.

More about Lan He & all articles ↗

Give them a reasonto stop by.

Send us your phone photos. Get posts ready to share.

See what we can make
Stillmore fashion creative look
All stories ↗