T
TENCENT / tencent/WeMM-Video-35BA3B

WeMM-Video-35BA3B

A Chinese-English bilingual family for understanding text, images, audio and video with its soundtrack. It returns text and offers direct-answer or reasoning modes. The 2B, 4B, 9B and 35B-A3B BF16 checkpoints were published on November 20; quantizations are not new models.

01 / SUMMARY

The essential

ORIENTATIONBilingual audiovisual understanding

Primary use stated or inferred cautiously from official documentation.

CONTEXT8,192 tokens in the deployment recipe

Maximum input capacity when published by the source.

EXITText; separate output maximum unpublished

Documented maximum generation limit.

ACCESSHugging Face · downloadable checkpoint

Verified availability channels.

MODEL TYPEMultimodal · Voice and audio · Video

Primary functional family and verified specialties.

LICENSEApache 2.0

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

02 / TECHNICAL SHEET

Limits and integration

API IDtencent/WeMM-Video-35BA3B
Model typeMultimodal · Voice and audio · Video
Access modelOpen weights
LicenseApache 2.0
DeploymentSelf-hosted
Release20 NOV 2026
Knowledge cutoffNot published
EntranceText · Image · Audio · Video
ExitText
Context window8,192 tokens in the deployment recipe
maximum outputText; separate output maximum unpublished
ReasoningNot published
Published tools
Structured OutingsNot published
Batch processingNot published
Prompt cacheNot published
Fine-tuningNot published
Verified platformsHugging Face · Transformers
BEST SUITED FOR

Describe scenes, transcribe speech and reason jointly about images and sound.

WORTH MONITORING

The recipe uses 8,192 tokens and up to four images, two videos and two audio items per request; this does not establish the architecture’s practical maximum. Custom code and runtime patches are required. The MoE variant’s 3B active parameters do not mean only 3B parameters need loading.

03 / CAPABILITIES

What it can do

01

Orientation

Describe scenes, transcribe speech and reason jointly about images and sound.

02

Context

8,192 tokens in the deployment recipe · Text; separate output maximum unpublished

03

Tools and integration

Not published

04

Access

Hugging Face · downloadable checkpoint

Input modalities
TextImageAudioVideo
04 / PRICES

Documented cost

ConceptWorthUnit/condition
Standard inputNo published official API rateStandard API rate
Cached inputNot publishedReading reused prefixes
Cache write or storageNot publishedThe condition varies by provider
Standard outputSelf-hosted infrastructure costMay include reasoning tokens
Batch inputNot publishedAsynchronous processing
Batch outputNot publishedAsynchronous processing

No verified official API tariff. Self-hosting requires memory, compute and maintenance; downloadable weights do not make inference free.

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

“

The 35B-A3B chart reports Video-MME 86.4, Daily-Omni 83.4 and AVHBench 80.0. These are manufacturer results under its protocols, not reproduced by Inferama. They do not transfer to 2B, 4B, 9B or quantizations.

BenchmarkResultMetricSource
Video-MME86.4Published scoreView source ↗
Daily-Omni83.4Published scoreView source ↗
AVHBench80.0Published scoreView 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

tencent/WeMM-Video-35BA3B

BF16 checkpoint; FP8, AWQ, GGUF and NVFP4 are formats of the same model.

07 / SOURCES

Traceability