T
TENCENT / hy4-flash

Tencent Hy4 Flash

Tencent multimodal model launched on November 16 for documents, interfaces and agents. It has 224 billion total parameters, including 64 billion Engram parameters, and activates 12 billion. It understands text and images and generates text; it is distinct from Hy4 and Hy4 Preview.

01 / SUMMARY

The essential

ORIENTATIONLow-cost visual agents

Primary use stated or inferred cautiously from official documentation.

CONTEXT1M tokens; maximum 960K input on TokenHub

Maximum input capacity when published by the source.

EXIT64K tokens on TokenHub

Documented maximum generation limit.

ACCESSTencent TokenHub

Verified availability channels.

MODEL TYPEMultimodal · Code · Text and reasoning

Primary functional family and verified specialties.

LICENSEProprietary

Declared terms for API access, model weights, or self-hosting.

02 / TECHNICAL SHEET

Limits and integration

API IDhy4-flash
Model typeMultimodal · Code · Text and reasoning
Access modelPaid proprietary
LicenseProprietary
DeploymentHosted API
Release16 NOV 2026
Knowledge cutoffNot published
EntranceText · Image
ExitText
Context window1M tokens; maximum 960K input on TokenHub
maximum output64K tokens on TokenHub
ReasoningNot published
Published toolsFeatures
Structured OutingsDocumented structured output
Batch processingNot published
Prompt cacheDocumented input caching
Fine-tuningNot published
Verified platformsTencent TokenHub · OpenRouter
BEST SUITED FOR

Document research, presentations, visual analysis and everyday tool-based tasks. TokenHub documents reasoning, functions, structured outputs and caching. Computer applications need an external environment and permissions.

WORTH MONITORING

A million-token context does not mean a million-token output. Open weights are announced for later: this profile verifies the available API without assigning a future weights license. Quotas and limits may vary between providers.

03 / CAPABILITIES

What it can do

01

Orientation

Document research, presentations, visual analysis and everyday tool-based tasks. TokenHub documents reasoning, functions, structured outputs and caching. Computer applications need an external environment and permissions.

02

Context

1M tokens; maximum 960K input on TokenHub · 64K tokens on TokenHub

03

Tools and integration

Features

04

Access

Tencent TokenHub

Input modalities
TextImage
04 / PRICES

Documented cost

ConceptWorthUnit/condition
Standard inputCNY 0.50 / 1M tokens before discountStandard API rate
Cached inputCNY 0.02 / 1M tokens before discountReading reused prefixes
Cache write or storageNot publishedThe condition varies by provider
Standard outputCNY 2 / 1M tokens before discountMay include reasoning tokens
Batch inputNot publishedAsynchronous processing
Batch outputNot publishedAsynchronous processing

TokenHub lists CNY 0.50 input, CNY 2 output and CNY 0.02 cache reads per million tokens. A 50% discount applies until November 30, 2026 at 12:00 Beijing time; list prices resume afterwards. These are neither Hy4 prices nor USD rates.

i Prices change and may depend on level, region or context length. Check the source before making a decision.

05 / EVALUATIONS

How to read the results

“

Tencent reports satisfaction of 2.79 out of 4 in a blind test with 85 internal WorkBuddy users, 259 tasks and 27 categories. This is an internal evaluation of the system and its environment, not an independent measurement or a guarantee for the model alone.

BenchmarkResultMetricSource
Hy4 Flash WorkBuddy blind satisfaction2.79 / 4Satisfaction · internal blind testView source ↗

Inferama only highlights a “best result” when the metric, test set, configuration, and date allow for an equivalent comparison. The supplier's figures are presented as claims from its own source.

06 / VERSIONS

Chronology

hy4-flash

API launch; the future weights publication is recorded separately.

07 / SOURCES

Traceability