Skip to content
Model infoModelsSnapshot: 15 Oct 2026

ByteDance / Seed Live Interpretation

Seed LiveInterpret 2.0

Named 2025 research release for bidirectional Chinese-English speech translation with live voice replication.

Speech & voiceTranscription
restrictedClosed weightsSpeech-to-speechSpeech-to-textTranslation
Official documentation (opens in a new tab)Reviewed 15 Oct 2026
9 of 22 specification fields collected
Context windowNot collectedAwaiting source review
Maximum outputNot collectedAwaiting source review
Input billingNot collectedByteDance Seed research
Output billingNot collectedNo inferred price

Specifications

Provider-documented facts. “Not collected” means we have not verified that field yet, not that the feature is unsupported.

Context / input limit
Not collected

Exact-version API specification not established by the research publication.

Maximum input tokens
Not collected

Exact-version API specification not established by the research publication.

Maximum output
Not collected

Exact-version API specification not established by the research publication.

Reasoning
Not collected

Exact-version API specification not established by the research publication.

Tool calling
Not collected

Exact-version API specification not established by the research publication.

Structured output
Not collected

Exact-version API specification not established by the research publication.

Streaming
SupportedSource for Streaming (opens in a new tab)

Simultaneous research capability. A specific streaming transport or SDK contract is not established for this release.

Batch processing
Not collected

Exact-version API specification not established by the research publication.

Fine-tuning
Not collected

Exact-version API specification not established by the research publication.

Knowledge cutoff
Not collected

Exact-version API specification not established by the research publication.

Release date
2025-07-24Source for Release date (opens in a new tab)

Date of the provider release announcement; the paper first appeared July 23.

Parameters
Not collected

Exact-version API specification not established by the research publication.

Architecture
End-to-end duplex speech understanding and generation; continual training, supervised fine-tuning and reinforcement learning.Source for Architecture (opens in a new tab)
Languages
Chinese, EnglishSource for Languages (opens in a new tab)

Bidirectional Chinese-English release; the broader rolling service language inventory is not inherited.

Context caching
Not collected

Exact-version API specification not established by the research publication.

Audio formats
Not collected

Exact-version API specification not established by the research publication.

Audio sample rates
Not collected

Exact-version API specification not established by the research publication.

Input modalities
AudioSource for Input modalities (opens in a new tab)
Output modalities
Text, AudioSource for Output modalities (opens in a new tab)
Translation modes
Full-duplex simultaneous speech-to-text and speech-to-speech translation.Source for Translation modes (opens in a new tab)
Voice cloning
Zero-shot voice replication from live conversation without pre-recorded samples.Source for Voice cloning (opens in a new tab)
Provider-reported latency
Provider reports average first-word latency of 2.21 s for speech-to-text and 2.53 s for speech-to-speech in its evaluation. Not an API latency SLA.Source for Provider-reported latency (opens in a new tab)

Access & pricing

Prices belong to an endpoint and billing tier. A provider listing does not guarantee access for every account or region.

ByteDance Seed research

Provider source (opens in a new tab)
API model ID
Not collected

No immutable public API selector established for this named research release.

Access
Research documentation and demonstrations. No verified immutable API selector or current entitlement.
Regions
Not established
Release identity
The July 24, 2025 announcement names the playground model Doubao - Simultaneous Interpretation 2.0.Source for Release identity (opens in a new tab)

Endpoint pricing has not been collected. Check the provider for applicable rates.

Data & enterprise terms

Retention, training use, regional processing and enterprise terms have not been summarized for this endpoint.

Caching

Caching can reuse earlier input processing. Availability, lifetime and billing depend on the endpoint. A cache hit still generates a new response.

01 / INITIAL INPUTProcess & retain

The system processes eligible input and retains reusable state.

02 / LATER REQUESTCheck available state

Reuse depends on matching input, session state and the cache policy.

03 / CACHE HITReuse & generate

Reusing state can reduce repeated work. Any charges follow the billing rules below.

ByteDance Seed research

Not verified

No version-specific prompt-cache contract or tariff established.

How it works
Not collected

No version-specific prompt-cache contract or tariff established.

Minimum reusable input
Not collected

No version-specific prompt-cache contract or tariff established.

Lifetime & refresh
Not collected

No version-specific prompt-cache contract or tariff established.

How to enable it
Not collected

No version-specific prompt-cache contract or tariff established.

Usage fields
Not collected

No version-specific prompt-cache contract or tariff established.

Billing rules
Not collected

No version-specific prompt-cache contract or tariff established.

Limits & data boundaries
Not collected

No version-specific prompt-cache contract or tariff established.

No verified cache rate for this endpoint. An empty rate is not a zero charge.

Caching reviewed 15 Oct 2026

What the measurements say

LiveInterpret 2.0 speech-to-text human score

ByteDance Seed / paper authors · Provider-reported research evaluation
74.8/100

Provider evaluation of bidirectional Chinese-English translation on RealSI; human-rated accuracy.

Evaluation source (opens in a new tab)

LiveInterpret 2.0 speech-to-speech human score

ByteDance Seed / paper authors · Provider-reported research evaluation
66.3/100

Provider evaluation includes translation, speech latency, rate, pronunciation and fluency; not directly comparable to the speech-to-text score.

Evaluation source (opens in a new tab)

LiveInterpret 2.0 first-word latency, speech-to-text

ByteDance Seed / paper authors · Provider-reported research evaluation
2.21seconds

Provider-reported evaluation mean, not a production latency guarantee.

Evaluation source (opens in a new tab)

LiveInterpret 2.0 first-word latency, speech-to-speech

ByteDance Seed / paper authors · Provider-reported research evaluation
2.53seconds

Provider-reported evaluation mean, not a production latency guarantee.

Evaluation source (opens in a new tab)

Latency and throughput depend on the host, request size, reasoning settings and load. There is no measured speed or global usage figure in this record.

The record behind the model

Why it is included

Provider explicitly names these developed model releases. Closed-weight classification is inferred from the commercial/research distribution and absence of matching checkpoints in the reviewed ByteDance and Seed inventories. A paper license is not a model-weight license; an explicit version-specific license remains unverified.

Classification evidence (opens in a new tab)

Lifecycle dates

Research paper first submitted23 Jul 2025Source (opens in a new tab)

Sources reviewed