Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Krisp / End-of-turn prediction

Turn-Taking v1

Historical audio-only end-of-turn predictor, generation 1, for deciding when a voice agent can respond.

Audio processing
historicalClosed weightsTurn detection
Official documentation (opens in a new tab)Reviewed 13 Oct 2026
9 of 20 specification fields collected
Audio frames100 ms prediction framesSee specification notes below
Languages1 listedFull language list in specifications
Input billingNot collectedKrisp · historical model documentation
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
Maximum output
Not collected
Input modalities
AudioSource for Input modalities (opens in a new tab)
Output modalities
End-of-turn probabilitySource for Output modalities (opens in a new tab)
Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
SupportedSource for Streaming (opens in a new tab)

Realtime frame or stream processing, not text-token streaming.

Batch processing
Not collected
Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Parameters
6.1 millionSource for Parameters (opens in a new tab)
Architecture
Not collected
Languages
EnglishSource for Languages (opens in a new tab)

Older turn-taking guide; multilingual support belongs to v3.

Context caching
Not collected
Native model options
krisp-viva-tt-v1Source for Native model options (opens in a new tab)

Published SDK model name or model filename, not a hosted LLM model selector.

Streaming audio chunks
100 ms prediction framesSource for Streaming audio chunks (opens in a new tab)

Prediction cadence; not an input duration cap. SDK adapter examples may accept smaller chunks.

Published model size
65 MBSource for Published model size (opens in a new tab)
Published training data
About 2,000 hours of conversational speech and 700,000 speaker turnsSource for Published training data (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.

Krisp · historical model documentation

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

No exact public checkpoint selector confirmed for this access route.

Access
Documented historical generation. Current access to the exact old artifact is unverified; contact Krisp. Current SDK pricing is not a historical inference price.
Regions
Customer-managed device or server deployment
Deployment mode
Licensed local inference; customer device/serverSource for Deployment mode (opens in a new tab)
Billing basis
Commercial SDK pricing by agreement; public developer tiers list no numeric model-level priceSource for Billing basis (opens in a new tab)

Meeting Assistant seat prices do not establish a developer inference tariff.

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

Data & enterprise terms

Historical generation; current SDK security documentation is linked for context. Exact historical processing and retention terms have not been collected.

Read the service policy (opens in a new tab)

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.

Krisp · historical model documentation

Not verified

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

How it works
Not collected

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

Minimum reusable input
Not collected

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

Lifetime & refresh
Not collected

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

How to enable it
Not collected

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

Usage fields
Not collected

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

Billing rules
Not collected

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

Limits & data boundaries
Not collected

Loading model weights once or reusing an audio session is not a documented discounted prompt-cache feature.

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

What the measurements say

No benchmark observations have been added for this model. We do not turn a vendor description into a performance score.

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

Krisp identifies its AI models as proprietary, licenses the software and restricts extraction, modification and redistribution. The commercial SDK distributes licensed .kef inference artifacts through its customer portal, not open-source model releases. Historical/current generations are included only where named in provider sources.

Classification evidence (opens in a new tab)

Sources reviewed