Skip to main content

Current model · Tencent

Hy4 flash benchmarks, pricing and specs

Hy4 flash is a Tencent model released on November 16, 2026 that takes text and image input and produces text. It costs $0.07 per 1M input tokens and $0.281 per 1M output tokens and accepts up to 1,048,576 tokens of context.

Per 1M input tokens
$0.07Listed today
Per 1M output tokens
$0.281Listed today
Context window
1.05M#12 of 71 current models

Data verified November 17, 2026 · Dataset 2026-11-17

Share this

Specs

Key facts

Provider
Tencent
Model family
Hy4
Release date
November 16, 2026
Status
Current
Access
API only
API model id
hy4-flash
Context window
1,048,576 tokens
Max output
65,536 tokens
Input
text, image
Output
text
Task families
text, multimodal
Declared capabilities
reasoning, agentic coding, image understanding, chart understanding

Charts

Against every current text model

Where Hy4 flash lands among the current models of its task, on every independent benchmark, on price and on context.

Where it stands

0 of 60 list a lower price per 1M input tokens. Each bar is the share of current models it beats.

Hy4 flash among the current text models, one row per benchmark, price and context window.

Price and context

Price per 1M input tokens

#1of 60

$0.07 · cheaper than 100%

Price per 1M output tokens

#2of 60

$0.281 · cheaper than 98%

Context window

#12of 71

1.05M tokens · longer than 60%

Price

Hy4 flash pricing

It is the cheapest of 60 current text models priced per 1M input tokens.

UnitPrice (USD)TierAppliesSource
1M input tokens$0.07standardCurrent
1M output tokens$0.281standardCurrent
1M input tokens$0.003cachedCurrent

Benchmarks

Independent benchmark results

Scores come from independent evaluators that ran the model themselves, pinned to a dated snapshot. Rank is among current models in the same snapshot. Tencent's own launch figures are in the next section, because they are not measured the same way.

No independent benchmark has published a result for Hy4 flash yet. It is still compared below on price, context and capabilities.

Provider results

Reported by Tencent

Tencent's own results for Hy4 flash, beside the rivals it chose, as its table lists them. Labs pick the benchmarks and run them their own way, so compare within one table only. Internal benchmarks and other evaluators' indexes are left out.

Hy4 flash launch table, November 16, 2026Tencent · 2026-11-17

BenchmarkHy4 flashHy4DeepSeek-V4.1-FlashQwen3.8-Flash-NextGLM-5.3-FlashGPT-6 Luna
DeepSWE v1.164.6%74.9%73.8%65.1%67.8%67.1%
ProgramBenchalmost solved16.0%30.0%29.0%22.0%13.0%6.0%
Terminal-Bench 4.0Claude Code harness19.2%44.4%40.0%29.5%35.4%13.3%
APEX-Agents v1.1pass@163.1%75.2%72.9%57.6%51.5%44.7%
AutomationBench v1.0.652.2%56.2%54.3%49.3%48.4%38.1%
HLE-Fullwith tools55.2%59.1%57.1%54.1%52.2%49.7%
OSWorld 2.1full, partial score68.1%78.9%68.4%58.8%64.0%55.7%
MMMU-Pro75.8%78.6%76.7%77.2%77.0%77.5%
ZeroBench mainwith tools, pass@545.8%55.4%16.5%41.7%35.3%33.6%

Tencent's appendix to the Hy4 flash announcement, every model at its highest available reasoning setting and run by Tencent. DeepSeek-V4.1-Flash is the 1017 build. Tencent's Hy4 table gives DeepSeek-V4.1-Flash 51.0% on ZeroBench, this one 16.5%. Left out here: Tencent's internal WorkIndex benchmarks, the cost per task in yuan and the Hy4 preview column.

Fit

Strengths, cautions and best uses

Strengths
  • Tencent's most cost-effective Hy model: 224B parameters, 64B of them Engram parameters, with 12B active per token, distilled from Hy4, with image input and a 1M-token context (up to 960K input and 64K output tokens on TokenHub)
  • $0.07 input, $0.281 output and $0.003 cached input per 1M tokens, against $0.834, $2.501 and $0.042 for Hy4; until noon on November 30, 2026 (Beijing time) Tencent Cloud TokenHub bills it at half price, and WorkBuddy offers it free
  • In Tencent's table it beats GPT-6 Luna on seven of nine public benchmarks, for example 63.1% against 44.7% on APEX-Agents v1.1 and 68.1% against 55.7% on OSWorld 2.1
  • Tencent reports decoding above 150 tokens per second and an average cost per task on its internal WorkIndex of a twentieth of Hy4's and an eighth of DeepSeek-V4.1-Flash's
Cautions
  • In Tencent's own table it trails Hy4 on all nine public benchmarks and DeepSeek-V4.1-Flash on eight, most on Terminal-Bench 4.0 (19.2% against 44.4% and 40.0%)
  • API only for now: Tencent plans open weights without giving a date, and recommends Hy4 instead for long-horizon software engineering
  • No independent benchmark score is recorded here yet; Tencent's other evidence is internal: its WorkIndex and a blind test by 85 of its own WorkBuddy users, in which satisfaction came close to Claude Haiku 5.5
Best for
  • High-volume everyday office work: deep search, documents, slides and simple front-end demos
  • Cheap computer-use and agent steps where speed and cost per task matter more than peak accuracy

From benchmark to production

Building on Hy4 flash?

Public scores rarely hold on your own data. We test it against the alternatives on your real tasks, then build the agent or pipeline around the model that wins.

AI Model Cost AuditLaunch price $100, regular price $150

Alternatives

Alternatives to Hy4 flash

The closest current text models by price. The cheapest model that scores as well or better is listed first; failing that, the cheapest one within 3% of the benchmark's scale.

On score and price

Hy4 flash beside the 4 alternatives listed below, on Arena Text and on blended price per 1M tokens.

Hy4 flash and its alternatives on score and price.

Arena Text

Measured November 11, 2026, higher is better

Hy4 flash
Not scored
Claude Haiku 5.5
1,421
GPT-4.1 nano
1,322
GPT-6 Luna
1,441
Muse Ember
Not scored

Blended price per 1M tokens

Listed today, lower is better

Hy4 flash
$0.123
Claude Haiku 5.5
$0.2
GPT-4.1 nano
$0.175
GPT-6 Luna
$0.2
Muse Ember
$0.2

FAQ

Hy4 flash FAQ

How much does Hy4 flash cost?

Hy4 flash costs $0.07 per 1M input tokens and $0.281 per 1M output tokens at the standard tier. It is the cheapest of 60 current text models priced per 1M input tokens. Source: Tencent, verified November 17, 2026.

What is the context window of Hy4 flash?

Hy4 flash accepts up to 1,048,576 tokens of context. It can return up to 65,536 tokens in one response. That ranks #12 of 71 current text models by context length.

What benchmark results has Tencent published for Hy4 flash?

In the Hy4 flash launch table (November 16, 2026), Tencent reports 64.6% on DeepSWE v1.1, 16.0% on ProgramBench (almost solved), 19.2% on Terminal-Bench 4.0 (Claude Code harness) and 63.1% on APEX-Agents v1.1 (pass@1). These are Tencent's own figures, beside the rivals it chose, not independent measurements.

Is Hy4 flash open source or API only?

Hy4 flash is available through Tencent's API only; its weights are not published.

What are the best alternatives to Hy4 flash?

The closest current alternatives by price and measured performance are Claude Haiku 5.5 (43% more per 1M input tokens, $0.10 vs $0.07), GPT-4.1 nano (43% more per 1M input tokens, $0.10 vs $0.07), GPT-6 Luna (43% more per 1M input tokens, $0.10 vs $0.07) and Muse Ember (43% more per 1M input tokens, $0.10 vs $0.07).

AI Model Brief

The AI models worth knowing, once a week

What shipped, how it scores on independent benchmarks, what it costs per million tokens and what was deprecated, from the data behind our AI model benchmarks.

One email a week. We send a confirmation link first, and every issue has a one-click unsubscribe.

We use your address only to send the brief. See our privacy policy.

See this week’s brief

Sources

Sources

All 291 sources behind the benchmarks, with their versions, licenses and notes, are in the source catalog.

Cite this page

Van Data Team. "Hy4 flash Benchmarks, Pricing and Specs." AI Model Benchmark Dataset 2026-11-17. https://vandatateam.com/ai-model-benchmarks/hy4-flash

Reuse the data: JSON · CSV · or explore every model in the AI model benchmark comparison.

Production evaluation

Deciding whether Hy4 flash fits your workload?

Public benchmarks narrow the shortlist. We test the finalists on your own tasks, data and cost limits, then build the routing and review controls to run the winner in production.

  • Evaluation cases and acceptance thresholds tied to real work
  • Model routing, fallbacks and cost controls