Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Smallest AI / Pulse

Pulse 2.0

English streaming transcription with integrated turn detection and optional acoustic emotion and gender estimates.

availableClosed weightsSpeech-to-text
Official documentation (opens in a new tab)Reviewed 13 Oct 2026
11 of 21 specification fields collected
Audio sample rates16,000 Hz recommendedSee specification notes below
Audio formatslinear16, linear32, alaw, mulaw, opus, ogg_opusSee specification notes below
Input billingSee ratesSmallest AI API
Output billingSee ratesPublished units and conditions below

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
Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
SupportedSource for Streaming (opens in a new tab)

Realtime transcription stream, not LLM token generation.

Batch processing
Not supportedSource for Batch processing (opens in a new tab)

Streaming-only model.

Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Not collected
Architecture
Shared Pulse encoder with LoRA adapter and classification heads for emotion/genderSource for Architecture (opens in a new tab)

Provider description does not disclose full architecture or parameter count.

Context caching
Not collected
Audio sample rates
16,000 Hz recommendedSource for Audio sample rates (opens in a new tab)
Audio formats
linear16, linear32, alaw, mulaw, opus, ogg_opusSource for Audio formats (opens in a new tab)
Speaker diarization
SupportedSource for Speaker diarization (opens in a new tab)

Speaker labels are optional API outputs, not identity verification.

emotionDetection
Six labels with confidence: neutral, happy, angry, sad, fear, disgustSource for emotionDetection (opens in a new tab)

Acoustic estimates, not direct observation of mental state.

Access & pricing

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

API model ID
pulse-2
Access
wss://api.smallest.ai/waves/v1/stt/live?model=pulse-2. Set language explicitly and choose a region supporting it.
Regions
Automatic routing to Mumbai, India or Oregon, USA; api.india.smallest.ai and api.us.smallest.ai pin a region. Language availability can differ by region.
API endpoint URL
wss://api.smallest.ai/waves/v1/stt/live?model=pulse-2Source for API endpoint URL (opens in a new tab)
Request quotas
Standard plan: 100 concurrent requestsSource for Request quotas (opens in a new tab)

Prices retain their published currencies and billing units below.

Published billing units · original currency · observed 13 Oct 2026
CategoryUsageRateUnit
Speechstandard model card1$0.005audio minute

Rate conditions

  1. Standard-plan rate in the model card. Enterprise pricing is custom.
Pricing source (opens in a new tab)

Data & enterprise terms

Content is retained unless an eligible Enterprise data policy is enabled. Options: no content storage, a 24-hour to 90-day window, or x-expire-content: true for seven-day expiry. On other plans the header is ignored. Billing/usage metadata remains; settings are not retroactive. TTS saved history is excluded from window expiry. Speech-to-speech and cloning are not expressly covered.

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.

Smallest AI API

pulse-2
Not verified

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

How it works
Not collected

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

Minimum reusable input
Not collected

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

Lifetime & refresh
Not collected

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

How to enable it
Not collected

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

Usage fields
Not collected

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

Billing rules
Not collected

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

Limits & data boundaries
Not collected

No discounted prompt-cache contract verified for this model/access route. Audio continuity and model-weight storage are separate features.

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

Smallest retains rights in its AI models and prohibits model extraction. Its model cards name Smallest as developer; Electron, Hydra and Lightning v3.1 explicitly use proprietary licenses. Commercial self-hosting requires licensed artifacts and a license proxy, not an open-weight grant.

Classification evidence (opens in a new tab)

Sources reviewed