Skip to content
Model infoModelsSnapshot: 15 Oct 2026

Sophea AI · KIEFER SA / Sophea ASR K1

Sophea ASR K1

October 2026 hosted preview for English and German prerecorded audio. The current documentation describes multi-model word-level fusion; access and production pricing are supplied on request.

Transcription
previewClosed weightsSpeech recognition
Official documentation (opens in a new tab)Reviewed 13 Oct 2026
14 of 26 specification fields collected
Audio durationNot collectedAwaiting source review
File size limit200 MB per recordingSee specification notes below
Input billingNot collectedKIEFER SA · Sophea hosted API preview
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
TextSource for Output modalities (opens in a new tab)
Reasoning
Not collected
Tool calling
Not collected
Structured output
Not collected
Streaming
Not collected
Batch processing
SupportedSource for Batch processing (opens in a new tab)

Synchronous whole-file POST; provider recommends concurrency for evaluation. No asynchronous batch-job queue or discounted batch plan is inferred.

Fine-tuning
Not collected
Knowledge cutoff
Not collected
Release date
Not collected
Parameters
Not collected
Architecture
Multiple ASR hypotheses fused word by wordSource for Architecture (opens in a new tab)

October description differs from the September two-model per-clip arbiter. Exact component identities, count and weights are not published in the current contract.

Languages
English (en), German (de)Source for Languages (opens in a new tab)

English is the default. The current API guide supersedes the older Greek/English marketing description; Greek is not listed as an accepted current language value.

Context caching
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

Maximum audio file
200 MB per recordingSource for Maximum audio file (opens in a new tab)
Audio duration limits
Not collected

Recordings longer than 40 seconds are segmented server-side. Forty seconds is not a maximum accepted recording length.

Streaming audio chunks
Recordings over 40 seconds are segmented server-sideSource for Streaming audio chunks (opens in a new tab)
Audio formats
WAV, FLAC, OGG, MP3, M4ASource for Audio formats (opens in a new tab)

Docs also say most other formats, without an exhaustive accepted-extension list.

Audio sample rates
Any sample rate according to the preview documentationSource for Audio sample rates (opens in a new tab)

Exact supported decoder range and resampling rules are not enumerated.

Audio channels
Mono or stereoSource for Audio channels (opens in a new tab)
Output formats
JSON text, model and duration_sSource for Output formats (opens in a new tab)
Provider-reported latency
Approximately one second for a short clipSource for Provider-reported latency (opens in a new tab)

Provider typical-latency statement with unspecified clip length/hardware; not a service-level guarantee.

Usage metering
Production pilot pricing on request. Researcher and leaderboard-maintainer evaluation keys are free.Source for Usage metering (opens in a new tab)

Free evaluation is restricted eligibility, not a zero-dollar production usage rate.

Model card updated
October 2026 English model update and German support; evaluation dated October 7Source for Model card updated (opens in a new tab)

An exact October model release date or immutable version selector is not published.

Access & pricing

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

KIEFER SA · Sophea hosted API preview

Provider source (opens in a new tab)
API model ID
sophea-asr-k1

Default service selector; asr-k1 is an accepted alias. This is not an immutable release pin.

Access
Request pilot access from KIEFER SA. POST {API_URL}/v1/transcribe as multipart/form-data, with the privately supplied base URL. Authenticate using X-API-Key or Authorization: Bearer. Fields: file, language=en|de and model=sophea-asr-k1 (default; asr-k1 alias also accepted).
Regions
Endpoint location supplied with pilot access; company address in Athens does not establish an API residency guarantee
Languages
English (en), German (de)Source for Languages (opens in a new tab)

English is the default. The current API guide supersedes the older Greek/English marketing description; Greek is not listed as an accepted current language value.

Maximum audio file
200 MB per recordingSource for Maximum audio file (opens in a new tab)
Streaming audio chunks
Recordings over 40 seconds are segmented server-sideSource for Streaming audio chunks (opens in a new tab)
Audio formats
WAV, FLAC, OGG, MP3, M4ASource for Audio formats (opens in a new tab)

Docs also say most other formats, without an exhaustive accepted-extension list.

Audio sample rates
Any sample rate according to the preview documentationSource for Audio sample rates (opens in a new tab)

Exact supported decoder range and resampling rules are not enumerated.

Audio channels
Mono or stereoSource for Audio channels (opens in a new tab)
Output formats
JSON text, model and duration_sSource for Output formats (opens in a new tab)
Batch processing
SupportedSource for Batch processing (opens in a new tab)

Synchronous whole-file POST; provider recommends concurrency for evaluation. No asynchronous batch-job queue or discounted batch plan is inferred.

API aliases
asr-k1Source for API aliases (opens in a new tab)
API request fields
file, language, modelSource for API request fields (opens in a new tab)
Concurrent requests
Provider recommends 16–32 concurrent requests for batch evaluationSource for Concurrent requests (opens in a new tab)

Recommendation, not a documented maximum or paid concurrency entitlement.

Speaker diarization
Not collected

No diarization field or structured speaker output appears in the current preview contract. Broader product-page copy is explicitly illustrative.

Word timestamps
Not collected

No word timestamp output is documented in this preview contract.

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

Data & enterprise terms

API documentation states audio is processed in memory and not stored. Exact log/metadata retention, training rights, subprocessors and deployment region are not specified there and require the pilot agreement. Broad enterprise and website claims are not substituted for the ASR contract.

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.

KIEFER SA · Sophea hosted API preview

sophea-asr-k1
Not verified

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

How it works
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

Minimum reusable input
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

Lifetime & refresh
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

How to enable it
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

Usage fields
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

Billing rules
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

Limits & data boundaries
Not collected

No model-specific prompt-cache contract verified. Server-side audio segmentation, speech buffers and reused decoders do not establish discounted token caching.

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

What the measurements say

October preview · AMI Cleaned

KIEFER SA / Sophea authors · provider-reported
6.29% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · Earnings22 Cleaned AA chunked

KIEFER SA / Sophea authors · provider-reported
5.57% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · Gigaspeech Cleaned

KIEFER SA / Sophea authors · provider-reported
7.96% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · LibriSpeech test-clean

KIEFER SA / Sophea authors · provider-reported
0.91% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · LibriSpeech test-other

KIEFER SA / Sophea authors · provider-reported
2.05% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · SPGISpeech

KIEFER SA / Sophea authors · provider-reported
2.11% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · VoxPopuli Cleaned AA

KIEFER SA / Sophea authors · provider-reported
3.34% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · Voice Arena Monsoon en_IN

KIEFER SA / Sophea authors · provider-reported
3.23% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. English dataset variant as named, not a generic all-language accuracy estimate. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · FLEURS German

KIEFER SA / Sophea authors · provider-reported
2.73% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. 862 German clips. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

Evaluation source (opens in a new tab)

October preview · Mozilla Common Voice German

KIEFER SA / Sophea authors · provider-reported
2.33% WER

Provider measurement dated October 7, 2026 using the Open ASR harness and language passed per request. 13,511 German clips. Independent maintainer rerun was requested; these are not presented as independently reproduced scores.

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

The official model catalog explicitly labels ASR K1 proprietary, distinct from its listed open-weight families. The current service is a hosted preview with private endpoint/key issuance. This entry represents the closed combined speech system, not ownership of its original public base architectures.

Classification evidence (opens in a new tab)

Lifecycle dates

Current revision evaluated by provider7 Oct 2026Source (opens in a new tab)

Sources reviewed