ASR K1 (preview)
sophea
Speech to textTranscribes a recording into words
- Type
- Closed
- Input
- None held
- Output
- None held
- Cached
- None held
We don't hold a list price for this model yet · hosted only — no weights published
Our take
Written Sep 17, 2026ASR K1 (preview) is a speech-to-text model with solid measured accuracy on English audio, better than most models on every condition we hold figures for. The catch is reach: we list no download and no host, so there is no route to it we can point you to today.
Reach for it on clearly recorded English speech, where it gets 1.2% of words wrong on read-aloud material, or on podcast and video audio, where 7.7% is better than most models on that condition. Skip it if you need a route to run it today, if you need speaker labelling, timestamps or streaming, or if you need the lowest error rate available.
The case for it
- 1.2% of words wrong on clean read-aloud recordings, better than most models on that condition, so presentations and well-miked single-speaker material come out close to clean.
- 7.7% of words wrong on podcasts and video, better than most on that condition, so an internet-audio backlog is workable without hand-correcting every line.
- 6% of words wrong on accented English and 7.3% on meeting recordings, both better than most on their conditions, so non-native speakers and room audio stay ahead of the field middle.
The case against it
- We list no download and no host, so there is nothing here to run or to call — the accuracy figures describe a model you cannot currently reach.
- 4.3% of words wrong on average across nine English test sets, against a field best of 3.6% and a middle of 5.2%: above the middle, not the lowest error rate available.
- Every accuracy figure is English, and no source we hold measures the second language the model lists, so its accuracy there is unverified.
How good is it?
A speech-to-text model for turning recordings of meetings, podcasts and accented speakers into written text.
- turning spoken English into written textOpen ASR WER · 9th of 76
- transcribing recordings of meetings in a roomRecorded meetings · 13th of 92
- transcribing speakers with a range of accentsAccented speech · 15th of 76
- transcribing podcasts and video audioPodcasts and video · 22nd of 92
- transcribing clear recordings of people reading aloudClean read speech · 23rd of 92
TranscriptionTurning speech into text4 of 5Open ASR WER · 9th of 76
95.7%
Misses roughly one word in 23, averaged over nine English test sets.
2
Stated by the leaderboard; we do not hold the list itself.
Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case; the placing beneath each rate is against every model measured on that set.
The figures above come from the Open ASR Leaderboard, an independent public test that runs every model on the same recordings. It is the only measurement of transcription quality we know of, so there are no other scores to show.
Each of these is the same transcription job on a different kind of recording, so together they say where it holds up and where it slips — not how closely it follows an instruction.
Every published score for this model8 scoresEvery figure we hold, from 8 boards, with who ran it and a link to the source — including the boards no rating above is built on.
Where to get it
We hold no priced listing for ASR K1 (preview).
There is no copy to download and no host in our price data, so sophea is where to look. We watch OpenRouter, the provider APIs we track and the LiteLLM price set; this version appears in none of them, which is a gap in what we collect rather than a statement about what sophea sells.
When we formed this view
Recent changes
What moved
first indexed by our pipelineEach date is the day we first saw the change, or the day the maker announced it.
What we do not know about this model yet
- Nothing we hold says whether an endpoint streams, so we do not show it either way.
- We don't hold a list price for this model yet — the gap is ours, not the lab's.
Licence and identifiers
What the licence allowsWe hold no licence record for this model. Inside are the identifiers you need to pull it — its Hugging Face repo where we have one, our slug and a machine-readable card.
Licence
We hold no licence row of this model's own. A source states its weights are not published, so the determination that governs it is the one for closed weights, API access only.
Identifiers
- Takes in, gives back
- Audio in, text out
- Catalogue slug
- sophea-asr-k1-preview