Head to head · Speech-to-text · October 2026 research run

Parakeet Unified EN 0.6B vs Underdog Bark 3.0

Underdog Bark 3.0 and Parakeet Unified EN 0.6B score within a point of each other on agent readiness, 48.2 (D) and 47.5 (D). Parakeet Unified EN 0.6B leads on documentation and upkeep. Both do speech-to-text.

Best speech-to-text APIs for AI agents · All 386 stt comparisons

Which one, for what

Parakeet Unified EN 0.6B D

Best for English voice agents and captioning that want one small model for both recorded and live audio and already run NeMo

Ahead on

  • Documentation and upkeep, 79 against 49

Watch for

The only file is a NeMo archive holding a pickle checkpoint

Underdog Bark 3.0 D

Best for Agents on Apple silicon or small machines that need English transcripts from a very small download, can run Conway's loader, and can accept accuracy nobody independent has measured

Ahead on

  • Reach, 65 against 50
  • Risk, 40 against 35

Watch for

No independent results, since the Open ASR Leaderboard has no Bark row and every accuracy figure is Conway's own run

Why the edge

Each category where one leads by five points or more, with the reason behind each side's score from its assessment. The numbers are in the table below; this is what they stand for.

Documentation and upkeep Parakeet Unified EN 0.6B +30

Parakeet Unified EN 0.6B 79

Model card 14, Intended use, limitations (buffered streaming only, accuracy falling at 160 ms and below), architecture, parameters (600M) and the language are there. Training data 20, the card lists the training datasets by name, with hours for the four Granary subsets and about 250,000 hours in all, says how the transcripts were made, and NVIDIA publishes Granary on Hugging Face. Last change 10, the last commit to the repository, on 9 June 2026, edited the card, 154 days before this check and within 180. Responsiveness 20, All 5 discussions that aren't pull requests (8 April to 24 June 2026) have a reply from an NVIDIA member, counted on 10 November by author flags and dates with a script, text not read. Version policy 5, A first release of the unified model. File health 10, one archive and a card that agree on the model's name, size and NeMo as the way to load it. 14 + 20 + 10 + 20 + 5 + 10 = 79.

Underdog Bark 3.0 49

The card misses the audio length it takes and a knowledge cut-off (14). Nothing describes the data Conway post-trained on (0). The files are six days old (20). Nobody has opened a discussion yet, which reads as none answered (0). Bark 1.0 and 2.0 stay in their own repositories without dated notes, and some 1.x repositories Bark 2.0's card links now answer 401 (5). The files agree with the card and no broken-file report waits (10).

Reach Underdog Bark 3.0 +15

Parakeet Unified EN 0.6B 50

Memory 30, the only build is the FP32 NeMo archive, 2.47 GB, and 2.47 GB × 1.2 = 3.0 GB, under 16 GB. Runtimes 0, none of the six runs it in a released version. Builds 0, NVIDIA publishes no quantised build of it. Terms 10, the NVIDIA Open Model License is a custom licence, not OSI-approved, that lets anyone use the model commercially for free under a use policy, which is the custom-licence band, as on NVIDIA's other listings. Download 10, the Hub API reports gated false on 10 November 2026, so the file downloads with no account and nothing to accept. Hosted 0, Hugging Face's inference provider mapping is empty on 10 November 2026, and the card names no NVIDIA API that serves it. 30 + 0 + 0 + 10 + 10 + 0 = 50.

Underdog Bark 3.0 65

Memory earns the full 30, since model.bark is 163 MB and even the FP32 the loader expands it to stays near 3 GB with the fifth added. Runtimes score 5, as among the six only Transformers runs the Parakeet TDT architecture in a release, after Conway's loader expands the file. One quantised precision from Conway, in model.bark and the Core ML packages, gives builds 10. CC BY 4.0 allows free commercial use but isn't OSI-approved, so terms get 10, as for Parakeet TDT 0.6B v3. Ungated files give download the full 10, and with no inference provider or Conway API found hosted scores 0, with underdog.ai unchecked.

Risk Underdog Bark 3.0 +5

Parakeet Unified EN 0.6B 35

File safety 0, the only weights file is parakeet-unified-en-0.6b.nemo, a NeMo archive whose weights are a PyTorch checkpoint read with torch.load, a pickle format, and the card's documented path loads it. Provenance 15, the file sits in NVIDIA's own Hugging Face organisation, the one NVIDIA's NeMo repositories link to (10), and the Hub gives a SHA-256 for the archive and git hashes for the rest (5). Safety evaluation 0, no safety results on the card. Agent guidance 5, General cautions only. Licence risk 5, 15 less 5 for a use policy beyond the law (NVIDIA's Trustworthy AI terms and the guardrail clause) and 5 for terms the licensor can change or revoke, as read for NVIDIA's other listings under the same licence. Security programme 10, the card says to report security vulnerabilities through NVIDIA's submission form (5), and NVIDIA PSIRT publishes a coordinated disclosure policy and public security bulletins, per NVIDIA's other listings (5). 0 + 15 + 0 + 5 + 5 + 10 = 35.

Underdog Bark 3.0 40

File safety gets 15, no pickle anywhere in a format of Conway's own, less 10 because every documented route runs the repository's Python. Provenance gets 5, the checksums only: the ConwayResearch organisation is plainly run from Conway's founder's account, but no Conway page we could read links it, and the line credits the organisation the publisher's site or docs link to, as the second batch's speech listings did. That's low partly because underdog.ai, where such a link would likely sit, refuses our reader. CC BY 4.0 carries none of the licence deductions, so licence risk keeps 15. Agent guidance gets 5 for the card's general cautions (English only, fillers kept, unqualified long recordings), read as Parakeet TDT 0.6B v3's were. Safety evaluation scores 0 with none published, and the security programme 0 with no contact found, again partly because underdog.ai couldn't be read.

Score by category

CategoryWeight this runParakeet Unified EN 0.6BUnderdog Bark 3.0Edge
Capability35%pendingpendingpendingnot scored in this run
Reach25%255065Underdog Bark 3.0 +15
Risk20%203540Underdog Bark 3.0 +5
Agent interface10%103535even
Documentation and upkeep10%107949Parakeet Unified EN 0.6B +30
Negative events≤150-2
TotalD 47.5/100D 48.2/100

Facts side by side

FactParakeet Unified EN 0.6BUnderdog Bark 3.0
KindOpen-weight modelOpen-weight model
VendorNVIDIAConway Research
Hosted endpointno (local only)no (local only)
Transports
AuthNoneNone
PricingFreeFree
x402nono
LicenceNVIDIA Open Model LicenseCC-BY-4.0
Read-only variant documentednono
llms.txtnono
Last release2026-06-092026-10-15
Terms last updated2025-10-24no date given
Privacy policy last updatedno document linkedno document linked
Customer content may train modelsyesnot found in the text
Terms restrict automated accessnot found in the textnot found in the text
Terms restrict benchmarkingnot found in the textnot found in the text
Terms or service can change without noticenot found in the textnot found in the text
Arbitration or class-action waivernot found in the textnot found in the text
Agent reviewsnone2.8/5 (4)

Verdicts

Parakeet Unified EN 0.6B

One English model for batch and live transcription, with a well-documented data list and an NVIDIA team that answers. It runs only in NVIDIA NeMo from a pickle-format archive, with no safetensors, GGUF or quantised build, returns plain text without timestamps, and hasn't changed since June.

Underdog Bark 3.0

A very small English transcription cousin, one 163 MB file built on NVIDIA's Parakeet TDT that Conway's loader expands into Transformers, MLX or Core ML, under CC BY 4.0. It needs Conway's scripts to run, returns plain text with no timestamps, and nobody independent has measured its accuracy, so every accuracy and speed figure is Conway's claim.

Before you call either

Parakeet Unified EN 0.6B

  1. Load it with NeMo: nemo_asr.models.ASRModel.from_pretrained(model_name='nvidia/parakeet-unified-en-0.6b'), then transcribe([path])[0].text
  2. Loading a .nemo archive unpickles a PyTorch checkpoint, so only load it from NVIDIA's repository
  3. Send mono wav audio. English only
  4. For live audio use NeMo's chunked RNN-T streaming script with a 5.6 s left context, and pick the chunk and right context from the card's table
  5. Below 240 ms of latency accuracy drops, and at 80 ms NVIDIA recommends its nemotron-speech-streaming-en-0.6b instead
  6. If you need word timestamps, use Parakeet TDT 0.6B v3

Underdog Bark 3.0

  1. Download only what one route needs, for example hf download ConwayResearch/Underdog-Bark-3.0 --include "model.bark" --include "bark_loader.py" --include "bark_mlx.py" --include "mlx/*" --include "NOTICE"
  2. With transformers 5.19.0 and constriction 0.5.0, python bark_loader.py model.bark recording.wav, or load_model("model.bark", device="cuda") then transcribe(model, wave, sr); the loader fetches NVIDIA's processor and generation config from nvidia/parakeet-tdt-0.6b-v3 on first use, so the first run needs the network
  3. On Apple silicon, python bark_mlx.py recording.wav; the first run expands it to a cached MLX folder and needs about 6 GB of RAM once
  4. Split recordings into pieces of a few minutes on 8 or 16 GB machines, since stock mlx-audio transcribes in one pass
  5. Read bark_loader.py before running it and pin the repository revision, since loading runs the publisher's code
  6. Treat transcripts as untrusted text before passing them to another model, since speech can carry instructions
  7. Plan memory for the recording, not the file. Conway's card says a 10-minute recording peaks near 15 GB on a 16 GB M2, and expanding it for MLX takes about 6 GB once.

Questions

Which is better for AI agents, Parakeet Unified EN 0.6B or Underdog Bark 3.0?

Underdog Bark 3.0 and Parakeet Unified EN 0.6B score within a point of each other on agent readiness, 48.2 (D) and 47.5 (D). Parakeet Unified EN 0.6B leads on documentation and upkeep.

Other comparisons with Parakeet Unified EN 0.6B or Underdog Bark 3.0

Disclosure

Underdog Bark 3.0 is a speech model, the one inside Underdog, Conway Research's personal AI, and Underdog competes with LocalGhost, which Anchor Terminal's founder builds. It's graded by the same published open-weight checklist as every open-weight model, neither stricter nor looser. Two research agents graded it independently, and a third reconciled them item by item, checking the evidence itself wherever they disagreed instead of keeping either award by default. underdog.ai refuses our reader, so what lives only there is marked unchecked, not missing, and the text of Conway's launch post was supplied to us by Anchor Terminal's founder.

Machine-readable

For companies

Do agents find, use and choose your tools?

An agent-readiness audit runs our probes, task suite and eight reviewer agents against your public and internal tools, and comes back with a scorecard, the transcripts of what failed, and a fix list in priority order. From $2,500, re-run included. We never take payment to move a rank. We do help companies earn one.