Project AURIONAI Use-Cases Research & Innovation
speech recognition

Canary-Qwen-2.5B

NVIDIA ASR model with Canary encoder and Qwen decoder; led the Open ASR Leaderboard in 2025.

NVIDIANVIDIAUSA
NVIDIA
Levelmodel
Classspeech recognition
Parameters2.5B
Contextnot in the notes
ReleasedJun 2025
Opennessopen weights permissive
Runs onsingle datacentre GPU
Statusactive

Benchmark Results

maintainer-run
Hugging Face Open ASR Leaderboardspeech
eight-set macro-average English WERmaintainer-run
5.63
048

Score in WER. Lower is better.

snapshot 2026-05-06 via benchmarklist6 May 2026 benchmarklist.com
Where this comes from
Every score names who ran it: vendor-reported, maintainer-run or independent. A rank is shown with the population it was ranked in (gap rule 3).

Lineage

finetune
Qwen to Canary-Qwen-2.5B: finetune, low confidenceCanary-1B to Canary-Qwen-2.5B: finetune, low confidenceQwenQwenAlibaba (Qwen)Canary-1BCanary-1BNVIDIA, 2024Canary-Qwen-2.5BCanary-Qwen-2.5BNVIDIA, Jun 2025
Alibaba (Qwen)Qwen is the parent (finetune)low confidencenotes
NVIDIACanary-1B is the parent (finetune)low confidencenotes
Where this comes from
Parent and child relations from model cards, Hugging Face base_model fields and the genealogy tables of the research notes. The tree shows two generations each way, 3 models. Confidence and evidence are the notes' own.

R&D Topics

Audio language models for speech understandingexemplar

Identity

Developer
NVIDIA
Family
not in the notes
Released
Jun 2025 sourcessecondary report
Weights published
–
Record checked
10 Oct 2026
Where this comes from
Developer and dates from the research notes. Release date is the vendor's public announcement; weights date is when open weights were published.

Architecture and Size

Architecture
autoregressive FastConformer encoder + Qwen LLM decoder (unverified detail)
Total parameters
2.5B sources
Active parameters
–
Checkpoint parameters
–
Context length
not in the notes
Input
audio
Output
text
Where this comes from
Parameter counts as the developer states them. Where the checkpoint holds a different total, both are shown (register A16).

Openness and Access

Openness
open weights permissive
Licence
Creative Commons Attribution 4.0 sources
Access routes
open weights download
Where this comes from
The licence file governs over registry tags and aggregator labels.

Deployment

Runs on
single datacentre GPU
Footprint
not in the notes
Runtimes
not in the notes
Where this comes from
Deployment tiers and runtimes as named in the notes. Footprint is the notes' figure, not a measurement by AURION.

Indonesia

Indonesian support
not stated
Regional languages
not in the notes
Jakarta hosting
unknown
Languages
not in the notes
Where this comes from
Indonesia fields are carried on every model; unknown is an allowed value.

Industry Use Cases

No industry use case links this model.

Sources

  1. Gladia - Best open-source speech-to-text models in 2026 gladia.io https://www.gladia.io/blog/best-open-source-speech-to-text-models
  2. Open ASR Leaderboard Benchmark Scores & AI Model Leaderboard | BenchmarkList benchmarklist.com https://benchmarklist.com/benchmarks/open_asr_leaderboard/
  3. Speech Recognition Accuracy Leaders, 2022–2026 select.keenable.ai https://select.keenable.ai/r/speech-recognition-accuracy-leaders-2022-2026/JbURQjlVooLQ6J10NkbRVw

To cite this page: Project AURION, AI Use-Cases Research & Innovation, "Canary-Qwen-2.5B", https://aurion.semai.id/models/nvidia-canary-qwen-2-5b/, records of October 2026.