Models / Z.ai/ GLM ASR Nano 2512

GLM ASR Nano 2512

Z.ai · released Dec 9, 2025 · zai-org/GLM-ASR-Nano-2512

Speech to textTranscribes a recording into words

Input: audio. Output: text.InputOutput
Type
Open weightsMIT License
Languages
2
Size
2.3B

Context measured in tokens

Our take

Written Aug 2, 2026

GLM ASR Nano 2512 is a tiny downloadable speech-to-text model from Z.ai that turns audio into written words at exceptional speed. It handles English and Chinese under a permissive MIT licence, though its accuracy drops sharply in noisy or accented conditions.

Who should pick it

Pick this for offline or edge transcription where speed matters most — it processes an hour of audio in about eleven seconds on benchmark hardware. Use it for clean read-aloud or scripted audio, or projects needing a permissive open licence. Skip it if you are transcribing meetings, accented speech, or podcasts and video, where its error rate rises more than eightfold; or if you need hosted inference, as no providers are currently listed.

The case for it

  • Extremely fast: processes an hour of audio in roughly eleven seconds on benchmark hardware.
  • Strong on clean read-aloud speech, with a word error rate roughly 3.6 times better than its own overall average.
  • Permissive MIT licence allows commercial use, modification and redistribution.
  • Small enough for edge deployment at 2 billion parameters.

The case against it

  • Accuracy collapses in challenging conditions: recorded meetings show an error rate more than eight times higher than clean read speech.
  • No hosted inference options currently listed in our data.
  • Overall word error rate of roughly one word in seventeen, with no comparison data to judge it leading for general use.
00

How good is it?

TranscriptionTurning speech into text2.5 of 5Open ASR WER · 42nd of 74

Words it gets right

94%

Misses roughly one word in 17, averaged over nine English test sets.

How fast it listens

333×45th of 62

an hour of audio in 11 seconds, on the board's own hardware. Your machine will differ.

Languages

2

Listed on the model card. The accuracy above is English only.

Where it struggles
Read aloudaudiobooks, clean recording1.7%
Podcasts and videoeveryday internet audio8%
Accented speechspeakers from many countries10.7%
Meetingsa room, several people, far microphone14%

Percentage of words wrong on each set, lower better. Bars are scaled to this model's own worst case, not to the board.

Which languages

English · Chinese

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.

Also scored, on boards we give no mark for
European-accented speech 7th of 74Financial calls 11th of 74Podcasts and video 29th of 74Accented speech 31st of 74Harder read speech 48th of 74Clean read speech 52nd of 74Recorded meetings 64th of 74

These tests check whether a model follows instructions — a precondition for all the work above, but not a measure of how well that work is done, which is why they get no rating.

Every published score for this model9 scoresEvery figure we hold, from 9 boards, with who ran it and a link to the source — including the boards no rating above is built on.
333independentsource ↗
6independentsource ↗
10.7independentsource ↗
1.9independentsource ↗
14independentsource ↗
1.7independentsource ↗
3.8independentsource ↗
01

Can you run it yourself?

Fits in memory
weights load entirely on the card
Spills to system RAM
some weights offload; much slower
Too large
will not load even with offload
est
size is calculated; the verdict could change by 10%

Comfortable fit

A card many people ownFits in memory

GeForce RTX 4090 · 24 GB

Weights at Q4_K_M1.5 / 24 GBest
Spare memory20 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed578 tok/sest

Room to spare. 20 GB spare means a 10% error in the size would not change the answer.

One step upFits in memory

GeForce RTX 5090 · 32 GB

Weights at Q4_K_M1.5 / 32 GBest
Spare memory28 GB spare
Usable context262Kwhat the spare memory holds; no published limit on record
Decode speed1028 tok/sest

Room to spare. 28 GB spare means a 10% error in the size would not change the answer.

On a MacFits in memory

Apple M2 (8-core GPU, 8GB unified) · 8 GB

Weights at Q4_K_M1.5 / 8 GBest
Spare memory3.2 GB spare
Usable context33Kwhat the spare memory holds; no published limit on record
Decode speed50 tok/sest

Room to spare. 3.2 GB spare means a 10% error in the size would not change the answer.

Your hardware
Checking your profile…

Memory use by level

Against a 24 GB card.

Q4_K_M
recommended
1.5 GBest
Fits in memory
Q5_K_M
1.7 GBest
Fits in memory
Q8_0
2.5 GBest
Fits in memory

All 3 sizes here are calculated, not measured. We hold no measured file for this model, so each size comes from the parameter count and every verdict above inherits that uncertainty.

Check against your own machine → · All 71 devices, with every size →

02

When we formed this view

Dates behind this page

Aug 2, 2026BenchmarkScored 10.7 on Accented speechleaderboard
Aug 2, 2026BenchmarkScored 1.9 on Financial callsleaderboard
Aug 2, 2026BenchmarkScored 14 on Recorded meetingsleaderboard
Aug 2, 2026BenchmarkScored 2.1 on European-accented speechleaderboard
Aug 2, 2026BenchmarkScored 8 on Podcasts and videoleaderboard
Aug 2, 2026BenchmarkScored 1.7 on Clean read speechleaderboard
Aug 2, 2026BenchmarkScored 3.8 on Harder read speechleaderboard
Aug 1, 2026ListedListed on LLMapfirst indexed by our pipeline
Aug 1, 2026BenchmarkScored 333 on Open ASR RTFxleaderboard
Aug 1, 2026BenchmarkScored 6 on Open ASR WERleaderboard

Prices last checked 6h ago

What we do not know about this model yet

  • We hold no measured file for it, so all 3 sizes on this page are calculated from the parameter count.
  • Nothing we hold says whether an endpoint streams, so we do not show it either way.
03

Licence and identifiers

What the licence allowsMIT License, what it allows commercially, and the identifiers you need to pull this model — its Hugging Face repo, our slug and a machine-readable card.

Licence

MIT License

permissiveCommercial use allowed

Fully permissive: do anything with attribution. No patent grant, unlike Apache-2.0.

Identifiers

Architecture
Dense
Modality record
audio->text
Catalogue slug
zai-org-glm-asr-nano-2512

Machine-readable model card (omc.json) →

Something wrong on this page? Tell us