TabbitBlog

HY4 Flash vs DeepSeek V4.1 Flash: Prices, Evidence, and Fit

Compare Tencent TokenHub rates, official model routes, and the small amount of public head-to-head evidence without treating benchmark claims as guarantees.

In this article
  1. Key takeaways
  2. Same-provider Tencent TokenHub rates
  3. Why token price and job cost differ
  4. What public task evidence says
  5. Choose a test by workload
  6. A fair comparison recipe
  7. Verdict

At Tencent TokenHub list rates, HY4 Flash costs half as much as Tencent-hosted DeepSeek V4.1 Flash 1017 for uncached input and output. Cache-hit input costs the same. Those rates help compare API usage, but they do not say which model completes your job more cheaply or reliably.

Tencent separately reports HY4 Flash used one-eighth the task cost of DeepSeek V4.1 Flash 1017 in its WorkIndex benchmark at Max tier and peak pricing. That is a vendor benchmark claim. One Reddit user also described a single WorkBuddy arithmetic task where both models answered correctly and DeepSeek finished sooner. Neither result predicts all prompts. See the HY4 Flash review and pricing breakdown for the evidence and billing assumptions.

Key takeaways

  • On Tencent TokenHub, list input/output rates for HY4 Flash are ¥0.50/¥2 per million tokens; Tencent's DeepSeek 1017 is ¥1/¥4 off-peak and ¥2/¥8 at peak.

  • The cache-hit rates are ¥0.02/M for Flash and ¥0.02/M off-peak or ¥0.04/M at peak for DeepSeek.

  • Tencent's 1/8 task-cost figure is specific to WorkIndex, Max, and peak pricing.

  • A public same-task report is one user's three-model arithmetic run, not a controlled benchmark.

  • Compare the exact endpoint, time window, token usage, retries, tool calls, and accepted output.

Same-provider Tencent TokenHub rates

Model and timingUncached input / 1MOutput / 1MCache hit / 1M
HY4 Flash list¥0.50¥2¥0.02
HY4 Flash promotion, through Nov 30, 2026 12:00 Beijing¥0.25¥1¥0.01
DeepSeek V4.1 Flash 1017 off-peak¥1¥4¥0.02
DeepSeek V4.1 Flash 1017 peak¥2¥8¥0.04

Tencent defines DeepSeek peak as Beijing weekdays 09:00–12:00 and 14:00–18:00; server receipt time determines the rate. The model rates and promotion cutoff come from Tencent's TokenHub tariff. This table holds provider constant. DeepSeek's separate official API price page lists USD rates for its own endpoint; do not combine those with Tencent CNY rates in one estimate.

Why token price and job cost differ

For a fixed request, estimate API cost as:

(uncached input × input rate + cache-hit input × cache-hit rate + output × output rate) ÷ 1,000,000

Then include retries, tools, and any other billed services. Cost per successful task also depends on whether the answer passes your acceptance check. TokenHub's model page and the DeepSeek V4.1 Flash guide describe their respective routes.

Tencent's release says HY4 Flash's WorkIndex task cost was one-eighth of DeepSeek V4.1 Flash 1017 under the stated Max-tier and peak-price conditions. Tencent does not provide a universal conversion from that result to every prompt's token bill. Read the result as one evaluation comparison, not as a promise that your workload will cost 87.5% less.

What public task evidence says

In a November 16 Reddit post, one user ran three short advertising arithmetic prompts in WorkBuddy. The author said all three answers were correct; DeepSeek took 7 seconds and HY4 Flash 19 seconds. The post does not provide raw prompts, token counts, provider logs, or actual API bills. It is a useful example of a latency tradeoff to test, not a robust performance ranking. Read the original report.

Plotly's DataAnalyticsBench report reports a one-run HY4 Flash score of 39/43 on private synthetic data. It is not a direct comparison against DeepSeek, so it cannot settle which model is better for analytics. For broader model context, see HY4 Flash vs HY4 and the single-model review.

Choose a test by workload

WorkloadStarting pointWhat to measure
Repeated short extraction or summariesHY4 Flash trialaccepted answers, input/output/cache tokens, retries
Time-sensitive short responsesTest bothmedian and tail latency on the same endpoint and time window
Long reasoning or coding taskTest both on a fixed task settool-loop completion, correctness, repair attempts
Cost-sensitive production workflowProvider-matched trialcost per accepted task, including failures
Local deploymentNeither until a supported route is confirmedweights, license, memory and serving requirements

Tencent says HY4 Flash supports up to 1M context and plans an open-weight release, without a release date in the launch post. DeepSeek's official API page lists deepseek-flash for V4.1 Flash 1017, 1M context, and up to 384K output. These are endpoint limits, not a guarantee that both models will behave the same with a long prompt. See DeepSeek browser use and Tencent HY4 in a browser.

A fair comparison recipe

Use the same prompt, source material, acceptance rubric, concurrency, and time window. Record the full model ID and provider because deepseek-flash and Tencent's direct 1017 route have different rate cards. Run repeated trials; report median and tail latency, actual usage fields, retries, and pass rate. Keep human review for consequential outputs.

For agent work, include the complete tool loop rather than only the first response. A community launch discussion asks whether models select tools reliably, but it does not provide a measured failure rate. You can compare browser workflows with the DeepSeek Harness plugin guide, Feishu Base example, Tabbit CLI setup, and first browser-agent task checklist.

Verdict

Try HY4 Flash first when Tencent's lower TokenHub input/output list rates fit a bounded workload and you can validate outputs. Test DeepSeek too when latency, established endpoint behavior, or output limits matter. Use Tencent's task-cost result as a benchmark-specific signal, and make the final choice from repeated, provider-matched runs on your own success criteria.

Tabbit Browser offers Hy4 Flash. Download Tabbit, then select Hy4 Flash from the model list in Chat to start using it.

Tabbit Browser

FAQ

Is HY4 Flash cheaper than DeepSeek V4.1 Flash?

At Tencent TokenHub list rates, Flash's uncached input and output rates are half DeepSeek's off-peak rates. Cache-hit input is the same price. This is a provider-specific token-rate comparison, not a task-cost result.

Does HY4 Flash cost one-eighth as much as DeepSeek?

Tencent reports one-eighth the WorkIndex task cost against DeepSeek V4.1 Flash 1017 at Max and peak pricing. That is a vendor-reported evaluation result, not a universal API bill ratio.

Which model is faster?

One Reddit user reported DeepSeek finishing a single arithmetic task in 7 seconds and HY4 Flash in 19 seconds. One unverified task is not a general latency benchmark.

Do both models have a one-million-token context?

Tencent lists HY4 Flash with up to 1M context, while DeepSeek's official API table lists 1M context for deepseek-flash. Context and output limits can depend on the endpoint.

Can I run HY4 Flash locally?

Tencent announced that open-weight release is planned but gave no release date in its November 16 announcement. Confirm that weights and a license are actually published before planning local inference.

Take the next step

Let Tabbit work alongside you.

Research across tabs, automate repetitive browser work, and keep every piece of context within reach.