APPSO reports that, in its WorkBuddy comparison using the same prompts and tool settings, Hy4 Flash was faster and used fewer tokens than DeepSeek V4.1 Flash. Its output for a 3D historical sailing-ship task was simpler, while DeepSeek took longer and included more detail. The article does not provide task logs, prompts, or complete run parameters sufficient for independent recalculation.
Tasks this report can inform: Examples of web prototyping, file organization, document format conversion, presentation creation, and multi-agent office collaboration in WorkBuddy.
Tasks these results should not be generalized to: All office tasks, long-running software engineering, different agents or tool environments, or a general ranking of the models' speed, quality, or cost.
Model versions covered: Hy4 Flash; the comparison model is called DeepSeek V4.1 Flash in the cases, while the cost section calls it DeepSeek-V4.1-Flash-1017.
Test environment or client: WorkBuddy; the client version, tool list, model API, and complete harness configuration were not disclosed.
Reasoning level and parameters: Not specified. The article says the prompts, tool permissions, and other settings were kept the same for the comparison tasks, but does not disclose their specific values.
APPSO says it prepared the same tasks in WorkBuddy for Hy4 Flash and DeepSeek V4.1 Flash, keeping prompts, tool permissions, and other settings consistent, and compared execution, deliverables, elapsed time, and token usage. The article does not explain task order, number of repetitions, timing start and end points, model snapshots, or scoring method.
The article shows an interactive 3D historical sailing-ship task and says it also tracked first-pass delivery times for 10 coding and office tasks. Office examples also include converting a PDF paper into Markdown, Word, and a bilingual Chinese-English presentation, as well as researching Fable 5.2 with the WorkBuddy expert team. The article does not disclose per-task comparison data for these examples.
APPSO says Hy4 Flash completed a basic prototype of an interactive sailing-ship webpage in half an hour. DeepSeek V4.1 Flash took 1 hour and 30 minutes; its result had more natural water movement and more detailed sails.
The article reports about 149,000 tokens used and 4 million tokens read from cache for Hy4 Flash. It estimates a cost of ¥0.23 at Hy4 Flash API rates. For DeepSeek, the article reports about 300,000 tokens used, 16 million tokens read from cache, and an API cost of about ¥1.21.
The original article published a sailing-ship demo page. At collection time, the page exposed its title, a Three.js r160 scene and camera view, time, sails, and sea-condition controls; the scene itself remained in a loading state, so the rendered result and interaction quality could not be verified from it. The demo page is a deliverable, not independent evidence of timing or cost records.
APPSO says the two models' final results were broadly similar across 10 coding and office tasks, while Hy4 Flash had a clearer advantage in first-pass delivery time. The article's chart does not include a per-task data table, so the time for each task cannot be recovered from the text.
Hy4 Flash office examples include converting a PDF into Markdown, Word, and a bilingual Chinese-English presentation, as well as using the WorkBuddy expert team for in-depth research on Fable 5.2. The article presents these workflow examples but gives no corresponding elapsed times, token counts, or per-task quality scores.
APPSO also shows a public demo page for a “spot the difference” game generated by Hy4 Flash. This is a single-model example, not comparison data against DeepSeek.
| Task / metric | Hy4 Flash | DeepSeek V4.1 Flash (as named in the original article) |
|---|---|---|
| Time for 3D historical sailing ship | About 30 minutes | About 90 minutes |
| Tokens used for 3D historical sailing ship | About 149,000 used, about 4 million read from cache | About 300,000 used, about 16 million read from cache |
| Cost for 3D historical sailing ship | About ¥0.23 (estimated using Hy4 Flash API pricing) | About ¥1.21 (API cost reported in the article) |
| 10 coding and office tasks | The article says first-pass delivery was faster; per-task figures not disclosed | The article says final results were broadly similar; per-task figures not disclosed |
Costs and token counts are recorded as stated in the original article. It does not say whether cached tokens are included in the corresponding usage totals, and does not disclose per-call details; the costs were not independently verified against billing records.
This is a first-hand APPSO product-experience report, based on cases run by an editor in WorkBuddy. It is not an official Tencent benchmark or a third-party retest using a public methodology.
The limited conclusion supported by the sailing-ship case is that Hy4 Flash was faster and had a lower estimated cost in this reported task, while DeepSeek's result had more detail. A single task cannot establish a general ranking of the models' quality or speed in web development.
The article's conclusion about “10 coding and office tasks” is not accompanied by per-task timings, raw results, or scoring rules. It also publishes no comparison data for the PDF/presentation and expert-team cases. These can serve as experience signals, but not as a verifiable benchmark table.
WorkBuddy Agent's tool calls, caching, runtime environment, and timing conventions can affect the results. The article does not provide complete prompts, model revisions, call logs, repeated runs, or cost details, so the reported times and costs cannot be independently reproduced.
Reproduction would require the same WorkBuddy version, exact snapshots of both models, original prompts, tool permissions and versions, as well as timing rules, task acceptance criteria, and complete call logs. APPSO has not published these materials. The public sailing-ship page can be used to inspect the deliverable; during this visit, the 3D scene did not finish loading, and the page contained no generation record or cost evidence. Collection checked only the specified original article and its directly published artifacts; missing data was not filled in from secondary reports.
Hy4 Flash