NVIDIA Parakeet-TDT
Open ASR Leader
NVIDIA's open speech-recognition family — Parakeet-TDT-0.6B currently tops the Hugging Face Open ASR Leaderboard at 6.05% word error rate. The Token-Duration Transducer architecture is cheap to deploy (single-GPU realtime) and licensed permissively via NVIDIA's NeMo stack.
Cost
Free
Open weights — self-host
How are Intelligence, Speed & Cost bucketed?
Intelligence and Speed buckets are percentile ranks on
Artificial Analysis. Cost buckets are fixed dollar
thresholds keyed off output-token price ($/M out).
Intelligence
- Top 1%≤ 1%
- Top 5%≤ 5%
- Top 10%≤ 10%
- Good≤ 25%
- Medium≤ 50%
- Below avg> 50%
Speed
- Top 1%≥ 345 tok/s
- Top 5%≥ 237 tok/s
- Top 10%≥ 196 tok/s
- Good≥ 146 tok/s
- Medium≥ 90 tok/s
- Slow< 90 tok/s
Cost
- Freeopen weights · self-host
- Low< $1 / M out
- Moderate$1–5 / M out
- High≥ $5 / M out
Why it matters
Replaced Whisper as the default open-source ASR for many production deployments. Combined with the Canary-Qwen-2.5B record (5.63% WER) released months later, NVIDIA owns the open ASR frontier in 2025–2026.
Core Capabilities
Audio
Speech, music, or other audio understanding/synthesis.
Generative
Produces images, video, audio, or other media.
Context Window
Context window not disclosed.
Availability
API
Not available
Product / App
Not available
Open Source
Released
Enterprise
—
Pricing Model
Free / self-host
Open weights — pay only for compute.
Self-host What it feels like
- Audio model from Nvidia — see the linked sources below for benchmark and review coverage
- Audio synthesis or transcription per the published model card
Best use cases
- Audio synthesis / transcription tasks per the model card
- See the model spec and sources block for benchmarked use cases
Tools to try
Not ideal for
- Tasks far outside the modalities listed in this model's spec
- Workflows where a more recent successor in the same family scores higher