{ "data": { "a": { "slug": "nemotron-3-5-transcribe-30b-a4b", "name": "Nemotron 3.5 Transcribe 30B-A4B", "vendor": "NVIDIA", "vendorUrl": "https://github.com/NVIDIA-NeMo/Speech", "kind": "weights", "category": "speech-to-text", "summary": "Nemotron 3.5 Transcribe is NVIDIA's open-weight speech model of 30B parameters, about 4.3B active, that transcribes and translates recordings up to an hour in 40 languages, with speaker tags and word timestamps. OpenMDW-1.1. The 4-bit build fits 32 GB.", "url": "https://www.anchorterminal.com/tools/nemotron-3-5-transcribe-30b-a4b", "markdownUrl": "https://www.anchorterminal.com/tools/nemotron-3-5-transcribe-30b-a4b.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/nemotron-3-5-transcribe-30b-a4b.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/nemotron-3-5-transcribe-30b-a4b.json", "dns": "nemotron-3-5-transcribe-30b-a4b.grade.anchorterminal.com", "repo": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B", "license": "OpenMDW-1.1", "transports": [], "packages": [ { "registry": "huggingface", "name": "nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B" }, { "registry": "huggingface", "name": "nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-FP8" }, { "registry": "huggingface", "name": "nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-NVFP4" } ], "auth": "none", "authNotes": "The weights download from Hugging Face with no account, token or terms to accept. The BF16, FP8 and NVFP4 repositories are not gated (checked 2026-11-10).", "pricing": "free", "pricingNotes": "Free to download and use under OpenMDW-1.1, a permissive licence with no use policy or threshold. Running it yourself costs the hardware: 21.6 GB at 4 bits (NVFP4), 34.7 GB in FP8 or 67.5 GB in BF16. DeepInfra serves it for those without a GPU (checked 2026-11-10).", "priceSummary": "Free", "where": "local", "x402": { "level": "no", "evidence": "Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-11-10).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-11-10" }, "docsUrl": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B", "capabilities": [ "speech.stt", "speech.batch", "speech.languages", "speech.diarisation", "speech.translation", "model.open-weights" ], "tags": [ "open-weights", "model", "speech-to-text", "asr", "multilingual", "translation", "diarization", "timestamps", "long-form", "moe", "reasoning", "openmdw", "safetensors", "fp8", "nvfp4", "self-hosted" ], "lastRelease": "2026-10-29", "graded": true, "anchor": { "graded": true, "score": 61, "grade": "C", "agentReady": false, "rank": 0, "ranked": false, "notRankedWhy": "An open-weight model, graded by the open-weight checklist and ranked among open-weight models, not against tools", "rankOf": 1518, "categoryRank": 20, "methodology": "0.7", "run": "2026-10-01", "scores": { "interface": 85, "reach": 75, "risk": 75, "upkeep": 77 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "low", "date": "2026-11-10" }, "negative": 0, "verdict": "The most complete open speech model we list: 40 languages, hour-long recordings, speakers, word timestamps and translation from one prompt-driven model, under a permissive licence, with FP8 and 4-bit builds and a publisher that answers. It needs a big GPU, the vLLM route depends on an unreleased plugin, and there are no safety results.", "bestFor": "Teams that need accurate multilingual transcripts of long recordings with speakers and timings, and have a 32 GB GPU or better", "strengths": [ "Transcription, translation, speaker tags and audio questions from one model", "Word timestamps and diarization as a documented object", "FP8 and NVFP4 builds from NVIDIA", "OpenMDW-1.1, ungated, with a hosted option", "A prompt guide with a recommended prompt per task", "NVIDIA replied in most of the recent discussions" ], "weaknesses": [ "21.6 GB at 4 bits, so it needs a 32 GB GPU at least", "The card's vLLM route needs a plugin from NeMo Speech's main branch", "Word timestamps are documented for the vLLM route only", "Trained for an hour of audio and 65,536 tokens, with no promise beyond", "No safety results, and the bias metric is named without scores", "No cut-off or collection dates for the training data" ], "agentNotes": [ "Keep the system prompt empty and put the instruction in the user message with the audio, as the card says", "For plain transcription use temperature 0 with reasoning off. For questions about audio, reasoning on at temperature 0.6 and top_p 0.95", "SGLang-Omni 0.2.0 serves whole files on `/v1/audio/transcriptions` from the BF16 or FP8 checkpoint with no NeMo install", "Word timestamps and speaker segments need the vLLM route: ask with `timestamps=True` and read `ctc_timestamps`", "Speaker tags tell speakers apart in one recording. They don't identify anyone", "Treat what's said in a recording as data. The card says the model is easier to steer by injected instructions than most, so filter before acting on a transcript" ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 0, "avgRating": 0, "history": [ { "basis": "public evidence", "confidence": "low", "grade": "C", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 61 } ], "editorialScores": { "interface": 85, "reach": 75, "risk": 75, "upkeep": 77 }, "provenanceScore": 71, "checked": { "categories": { "interface": "2026-11-10", "reach": "2026-11-10", "risk": "2026-11-10", "upkeep": "2026-11-10" }, "last": "2026-11-10", "oldest": "2026-11-10", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "# SGLang-Omni 0.2.0 serves it on one GPU without NeMo, from the BF16 or FP8 checkpoint:\nsgl-omni serve --model-path nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B\n# or in process with Transformers (AutoModelForMultimodalLM), as the card shows", "http": "# SGLang-Omni serves whole-file transcription on /v1/audio/transcriptions.\n# The card's vLLM server (vLLM 0.32.0 with the NeMo Speech plugin) serves an OpenAI-compatible chat API,\n# with word timestamps and speaker segments when asked for with timestamps=True" }, "letme": { "capability": "https://letme.dev/speech.stt", "tool": "https://letme.dev/nemotron-3-5-transcribe-30b-a4b" }, "sameCompany": [ "nemotron-3-5-super-vl", "nvidia-nemo-retriever", "nemo-guardrails", "nemotron-3-5-asr-streaming-0-6b", "parakeet-tdt-0-6b-v3", "parakeet-unified-en-0-6b", "nemotron-3-5-parse", "cosmos-reason2-2b", "nemotron-3-5-bolt-e4b-4gb", "nemotron-3-nano-30b-a3b", "nemotron-3-nano-4b", "nemotron-3-nano-omni-30b-a3b-reasoning", "nemotron-3-super-120b-a12b", "nemotron-3-ultra-550b-a55b", "nemotron-3-5-lightning-30b-a3b" ], "weights": { "trainedBy": "NVIDIA", "builtOn": "NVIDIA Nemotron 3.5 Lightning 30B-A3B, per the card and the metadata", "params": "30B", "active": "4.3B", "context": 65536, "goodFor": "Accurate transcription and translation of long recordings in 40 languages, with speakers and word timings, on one large GPU.", "runtimes": [ "llama.cpp", "SGLang", "Transformers" ], "builds": [ { "name": "NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-NVFP4", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-NVFP4", "who": "publisher", "precision": "nvfp4", "sizeGB": 21.56, "memoryGB": 32, "format": "safetensors", "runsOn": [ "Transformers" ], "note": "NVIDIA's 4-bit build. The card lists it for Blackwell GPUs, down to an RTX 5090" }, { "name": "NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-FP8", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-FP8", "who": "publisher", "precision": "fp8", "sizeGB": 34.67, "memoryGB": 48, "format": "safetensors", "runsOn": [ "SGLang", "Transformers" ], "note": "NVIDIA's FP8 build. SGLang-Omni serves it" }, { "name": "NVIDIA-Nemotron-3.5-Transcribe-30B-A4B (BF16)", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B", "who": "publisher", "precision": "bf16", "sizeGB": 67.5, "memoryGB": 96, "format": "safetensors", "runsOn": [ "llama.cpp", "SGLang", "Transformers" ], "note": "The reference checkpoint, for accuracy and fine-tuning. The one layout llama.cpp converts" } ], "run": "sgl-omni serve --model-path nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-FP8", "runBuild": "NVIDIA-Nemotron-3.5-Transcribe-30B-A4B-FP8", "checked": "2026-11-10", "minMemoryGB": 32 }, "family": { "lab": "NVIDIA (Nemotron)", "family": "Nemotron 3", "url": "https://www.anchorterminal.com/open-weights/#nvidia", "others": [ { "name": "Nemotron 3 Nano 30B-A3B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16", "params": "30B total, 3B active (31.6B per the safetensors index)", "listing": "https://www.anchorterminal.com/tools/nemotron-3-nano-30b-a3b" }, { "name": "Nemotron 3 Nano 30B-A3B Base", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-Base-BF16", "params": "30B total, 3B active" }, { "name": "Nemotron 3 Super 120B-A12B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-BF16", "params": "120B total, 12B active", "listing": "https://www.anchorterminal.com/tools/nemotron-3-super-120b-a12b" }, { "name": "Nemotron 3 Super 120B-A12B Base", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Super-120B-A12B-Base-BF16", "params": "120B total, 12B active" }, { "name": "Nemotron 3 Nano 4B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Nano-4B-BF16", "params": "4B", "listing": "https://www.anchorterminal.com/tools/nemotron-3-nano-4b" }, { "name": "Nemotron 3 Nano Omni 30B-A3B Reasoning", "url": "https://huggingface.co/nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16", "params": "30B total, 3B active", "listing": "https://www.anchorterminal.com/tools/nemotron-3-nano-omni-30b-a3b-reasoning" }, { "name": "Nemotron 3 Ultra 550B-A55B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-BF16", "params": "550B total, 55B active", "listing": "https://www.anchorterminal.com/tools/nemotron-3-ultra-550b-a55b" }, { "name": "Nemotron 3 Ultra 550B-A55B Base", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-Base-BF16", "params": "550B total, 55B active" }, { "name": "Nemotron 3 Ultra 550B-A55B GenRM", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3-Ultra-550B-A55B-GenRM", "params": "550B total, 55B active" }, { "name": "Nemotron 3.5 Lightning 30B-A3B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-BF16", "params": "30B total, 3B active", "listing": "https://www.anchorterminal.com/tools/nemotron-3-5-lightning-30b-a3b" }, { "name": "Nemotron 3.5 Lightning 30B-A3B Base", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Lightning-30B-A3B-Base-BF16", "params": "30B total, 3B active" }, { "name": "Nemotron 3.5 Super VL 120B-A12B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-BF16", "params": "120B total, 12B active", "listing": "https://www.anchorterminal.com/tools/nemotron-3-5-super-vl" }, { "name": "Nemotron 3.5 Super VL 120B-A12B Base", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Super-VL-120B-A12B-Base-BF16", "params": "120B total, 12B active" }, { "name": "Nemotron 3.5 Bolt 74B-A9B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-74B-A9B-BF16", "params": "74B total, 9B active" }, { "name": "Nemotron 3.5 Bolt E4B 4GB", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Bolt-E4B-4GB", "params": "8B total, 4B effective", "listing": "https://www.anchorterminal.com/tools/nemotron-3-5-bolt-e4b-4gb" }, { "name": "Nemotron Labs 3 Elastic 30B-A3B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-3-Elastic-30B-A3B-BF16", "params": "30B total, 3B active" }, { "name": "Nemotron Labs 3 Puzzle 75B-A9B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-3-Puzzle-75B-A9B-BF16", "params": "75B total, 9B active" }, { "name": "Nemotron 3 Labs Ultra Math SFT", "url": "https://huggingface.co/nvidia/Nemotron-3-Labs-Ultra-Math-SFT", "params": "550B total, 55B active" }, { "name": "Nemotron 3 Labs Ultra Math RL", "url": "https://huggingface.co/nvidia/Nemotron-3-Labs-Ultra-Math-RL", "params": "550B total, 55B active" }, { "name": "Nemotron Labs 3 Competitive Coding 550B-A55B", "url": "https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-3-Competitive-Coding-550B-A55B-NVFP4", "params": "550B total, 55B active" }, { "name": "nemotron-3.5-asr-streaming-0.6b", "url": "https://huggingface.co/nvidia/nemotron-3.5-asr-streaming-0.6b", "params": "0.6B (638M per the safetensors count)", "listing": "https://www.anchorterminal.com/tools/nemotron-3-5-asr-streaming-0-6b" }, { "name": "Nemotron 3.5 Parse", "url": "https://huggingface.co/nvidia/Nemotron-3.5-Parse", "params": "0.631B (630,800,496 trainable parameters per the card)", "listing": "https://www.anchorterminal.com/tools/nemotron-3-5-parse" } ], "olderCount": 47 }, "area": "voice", "provenance": { "legalEntity": "NVIDIA Corporation", "domain": "nvidia.com", "domainRegistered": "1993-04-20", "endpointOnVendorDomain": null, "terms": "https://openmdw.ai/license/1-1/", "privacy": "", "statusPage": "", "changelog": "https://huggingface.co/nvidia/NVIDIA-Nemotron-3.5-Transcribe-30B-A4B/commits/main", "securityTxt": "none", "checked": "2026-11-10", "notes": [ "The model sits in NVIDIA's Hugging Face organisation and the card's governing terms are OpenMDW-1.1. The domain registration date is carried from the RDAP check made for nemotron-3-nano-30b-a3b on 21 October 2026.", "Nothing is hosted under this listing, so the licence stands in for terms of service and no privacy policy is scored.", "The changelog is the commit history, 13 commits from 27 to 29 October 2026.", "security.txt is carried from our other NVIDIA listings and wasn't re-read." ], "score": 71 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/nemotron-3-5-transcribe-30b-a4b.json" }, "answer": "Nemotron 3.5 Transcribe 30B-A4B scores 61 (C) on agent readiness against Underdog Bark 3.0's 48.2 (D), and leads in every scored category.", "b": { "slug": "underdog-bark-3", "name": "Underdog Bark 3.0", "vendor": "Conway Research", "vendorUrl": "https://conway.tech", "kind": "weights", "category": "speech-to-text", "summary": "Underdog Bark 3.0 is Conway Research's open-weight English speech recognition model, a post-trained and compressed Parakeet TDT 0.6B v3 in one 163 MB file, run with Conway's loader on Transformers, MLX or Core ML, under CC BY 4.0.", "url": "https://www.anchorterminal.com/tools/underdog-bark-3", "markdownUrl": "https://www.anchorterminal.com/tools/underdog-bark-3.md", "slimMarkdownUrl": "https://www.anchorterminal.com/tools/underdog-bark-3.min.md", "jsonUrl": "https://www.anchorterminal.com/api/v1/tools/underdog-bark-3.json", "dns": "underdog-bark-3.grade.anchorterminal.com", "repo": "https://huggingface.co/ConwayResearch/Underdog-Bark-3.0", "license": "CC-BY-4.0", "transports": [], "packages": [ { "registry": "huggingface", "name": "ConwayResearch/Underdog-Bark-3.0" } ], "auth": "none", "authNotes": "The files download from Hugging Face with no account, token or terms to accept. The repository isn't gated (checked 2026-10-21). The card's optional line about `hf auth login` applies only if access ever needs it.", "pricing": "free", "pricingNotes": "Free to download and use, commercial use included, under CC BY 4.0, which asks for attribution of NVIDIA, Moondream and Conway (the repository's NOTICE file). Running it costs the hardware. model.bark is 163.4 MB, and the Core ML packages about 0.2 GB each (checked 2026-10-21).", "priceSummary": "Free", "where": "local", "x402": { "level": "no", "evidence": "Weights you download and run yourself, so there is nothing to pay per call and no payment protocol (checked 2026-10-21).", "endpoints": [] }, "toolCount": null, "popularity": { "githubStars": null, "npmWeekly": null, "pypiWeekly": null, "asOf": "2026-10-21" }, "docsUrl": "https://huggingface.co/ConwayResearch/Underdog-Bark-3.0", "capabilities": [ "speech.stt", "speech.batch", "model.open-weights" ], "tags": [ "open-weights", "model", "speech-to-text", "asr", "english", "fastconformer", "tdt", "parakeet", "low-bit", "mlx", "core-ml", "apple-silicon", "on-device", "cc-by-4.0", "self-hosted" ], "lastRelease": "2026-10-15", "graded": true, "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.", "competesWith": "localghost", "anchor": { "graded": true, "score": 48.2, "grade": "D", "agentReady": false, "rank": 0, "ranked": false, "notRankedWhy": "An open-weight model, graded by the open-weight checklist and ranked among open-weight models, not against tools", "rankOf": 1518, "categoryRank": 30, "methodology": "0.7", "run": "2026-10-01", "scores": { "interface": 35, "reach": 65, "risk": 40, "upkeep": 49 }, "pending": [ "performance", "tasks" ], "assessment": { "confidence": "low", "date": "2026-10-21" }, "negative": -2, "negativeNotes": [ "A misleading claim, 2 off. Conway's launch post of 21 October 2026, whose text Anchor Terminal's founder supplied because underdog.ai refuses our reader, calls Underdog Bark 3.0 'the #1 frontier speech model'. The Open ASR Leaderboard shows that false on every reading the post's own words allow, even granting Conway's own 4.937 per cent average over the seven public English sets its card uses, which isn't on the leaderboard. As the best speech model, 32 of the leaderboard's 65 rows average lower on those sets. As the best open model, 16 open-weight rows do, led by Hojo-ASR-V1 at 4.291. On AMI, the set the post cites, 19 rows score below its 7.82. And at its own size in the leaderboard's size column, two open models do better, Parakeet TDT 0.6B v2 (0.6B, 4.819) and Granite Speech 5.0 470M NC (0.47B, 4.921). Conway's card claims only state of the art for its size, measured by download, which the leaderboard can't test and is recorded as Conway's claim. 2, the low end of the range, because the card states the narrower position and the post reached us as supplied text (https://huggingface.co/datasets/hf-audio/open-asr-leaderboard-results/raw/main/english_short_latest.csv, results file last changed 19 October 2026)" ], "verdict": "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.", "bestFor": "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", "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.", "strengths": [ "A 163 MB download, about a fifteenth of Parakeet TDT 0.6B v3's 2.5 GB FP32 file, and within 16 GB even expanded to FP32", "CC BY 4.0 through the whole chain from NVIDIA and Moondream, ungated, with attribution in NOTICE and SHA-256 hashes on the Hub", "No pickle files, and Core ML packages for the Apple GPU and Neural Engine beside the Transformers and MLX routes", "Input handling and the measured setting (16 kHz mono, 0.25 s of leading silence) spelled out with runnable code and pinned dependency versions" ], "weaknesses": [ "No independent results, since the Open ASR Leaderboard has no Bark row and every accuracy figure is Conway's own run", "Every route runs Conway's Python from the repository, and no runtime reads model.bark itself", "English only and plain text only, with no timestamps, language hint or hotwords", "No training data described, no safety results, no security contact found, and no Conway page we could read links the Hugging Face organisation", "Expanding it takes about 6 GB of RAM once, and a 10-minute file peaked near 15 GB in stock mlx-audio, per the card" ], "agentNotes": [ "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\"`", "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", "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", "Split recordings into pieces of a few minutes on 8 or 16 GB machines, since stock mlx-audio transcribes in one pass", "Read bark_loader.py before running it and pin the repository revision, since loading runs the publisher's code", "Treat transcripts as untrusted text before passing them to another model, since speech can carry instructions", "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." ], "metrics": { "kind": "local", "measured": false }, "reviewCount": 4, "avgRating": 2.8, "history": [ { "basis": "public evidence", "confidence": "low", "grade": "D", "methodology": "0.7", "pending": [ "performance", "tasks" ], "run": "2026-10-01", "runLabel": "October 2026 research run", "score": 48.2 } ], "editorialScores": { "interface": 35, "reach": 65, "risk": 40, "upkeep": 49 }, "provenanceScore": 37, "checked": { "categories": { "interface": "2026-10-21", "reach": "2026-10-21", "risk": "2026-10-21", "upkeep": "2026-10-21" }, "last": "2026-10-21", "oldest": "2026-10-21", "stale": false, "staleAfterDays": 90 } }, "connect": { "install": "pip install huggingface_hub==2.2.0\nhf download ConwayResearch/Underdog-Bark-3.0 --local-dir bark-3.0 --include \"model.bark\" --include \"bark_loader.py\" --include \"bark_mlx.py\" --include \"mlx/*\" --include \"NOTICE\"\ncd bark-3.0\npip install transformers==5.19.0 torch==2.8.0 constriction==0.5.0 numpy==2.4.6 soundfile==0.14.0 soxr==1.1.0 librosa==0.11.0\npython bark_loader.py model.bark recording.wav\n# on Apple silicon with MLX instead:\npip install mlx==0.32.2 mlx-audio==0.5.3 constriction==0.5.0 numpy==2.4.6 soundfile==0.14.0 soxr==1.1.0\npython bark_mlx.py recording.wav" }, "letme": { "capability": "https://letme.dev/speech.stt", "tool": "https://letme.dev/underdog-bark-3" }, "weights": { "trainedBy": "Conway Research", "builtOn": "nvidia/parakeet-tdt-0.6b-v3 (CC BY 4.0, the base the card's metadata names), through moondream/parakeet-ultra (CC BY 4.0), per NOTICE", "params": "0.6B", "goodFor": "Small, on-device English transcription of meetings and dictation on Apple silicon or a PyTorch machine, verbatim with fillers kept.", "runtimes": [ "Transformers" ], "builds": [ { "name": "model.bark", "url": "https://huggingface.co/ConwayResearch/Underdog-Bark-3.0/blob/main/model.bark", "who": "publisher", "precision": "low-bit (entropy-coded per-row levels for encoder linear weights, 5 or 6 bit integers for other matrices, fp16 small tensors)", "sizeGB": 0.16, "memoryGB": 8, "format": "bark", "runsOn": [ "Transformers", "mlx-audio" ], "note": "Conway's own format, read only by the repository's bark_loader.py, which expands it to FP32 for Transformers, or by bark_mlx.py, which writes an FP16 or 4-bit MLX folder for mlx-audio. About 2.1 bits a parameter as stored, entropy-coded; the bit width isn't stated The band comes from the file size, as for every build. Conway's card says expanding it for MLX takes about 6 GB once and a 10-minute recording peaks near 15 GB on a 16 GB M2, so long recordings need more than the band." }, { "name": "Core ML (Apple GPU)", "url": "https://huggingface.co/ConwayResearch/Underdog-Bark-3.0/tree/main/coreml/gpu", "who": "publisher", "precision": "the same quantised values as model.bark, FP16 encoder arithmetic", "sizeGB": 0.2, "memoryGB": 8, "format": "coreml", "runsOn": [ "Core ML" ], "note": "BarkEncoder.mlpackage with a fixed 15-second window, decoder.npz and vocab.json, run by coreml/gpu/bark_coreml.py" }, { "name": "Core ML (Apple Neural Engine)", "url": "https://huggingface.co/ConwayResearch/Underdog-Bark-3.0/tree/main/coreml/ane", "who": "publisher", "precision": "the same quantised values as model.bark, FP16 encoder arithmetic", "sizeGB": 0.2, "memoryGB": 8, "format": "coreml", "runsOn": [ "Core ML" ], "note": "BarkEncoder.mlpackage with 8 and 15 second functions sharing weights, decoder.npz and vocab.json, run by coreml/ane/bark_coreml.py" } ], "run": "python bark_loader.py model.bark recording.wav", "runBuild": "model.bark", "checked": "2026-10-21", "minMemoryGB": 8 }, "family": { "lab": "Conway Research", "family": "Underdog Bark", "url": "https://www.anchorterminal.com/open-weights/#conway" }, "area": "voice", "provenance": { "legalEntity": "", "domain": "conway.tech", "domainRegistered": "2025-11-08", "endpointOnVendorDomain": null, "terms": "https://creativecommons.org/licenses/by/4.0/legalcode.en", "privacy": "", "statusPage": "", "changelog": "https://huggingface.co/ConwayResearch/Underdog-Bark-3.0/commits/main", "securityTxt": "none", "checked": "2026-10-21", "notes": [ "The model sits in the ConwayResearch organisation on Hugging Face, named Conway and unverified, whose Bark 3.0 commits both come from the Hugging Face account of Sigil Wen, whom sigilwen.ca/conway names as Conway's founder. conway.tech (page and script), sigilwen.ca/conway and Conway-Research/automaton's README don't link the organisation, and underdog.ai, which the card links, refuses our reader, so a link from Conway's own site is unchecked.", "No registered company name found, as in the Underdog listing of 3 October 2026. The Hugging Face organisation and conway.tech both say Conway. RDAP gives conway.tech a registration date of 8 November 2025 (rechecked 21 October 2026).", "Nothing is hosted for this listing, so the licence stands in for terms of service and no privacy policy is scored.", "The changelog is the commit history, two commits on 15 October 2026 (initial commit and Underdog Bark 3.0). No release notes outside underdog.ai were found.", "conway.tech/.well-known/security.txt answered 404 on 21 October 2026, and the repository has no security file. underdog.ai is unchecked." ], "score": 37 }, "pageJsonUrl": "https://www.anchorterminal.com/tools/underdog-bark-3.json", "live": { "slug": "underdog-bark-3", "securityTxt": { "url": "https://conway.tech/.well-known/security.txt", "state": "none", "checkedAt": "2026-11-10T16:07:53.863957393Z" }, "domain": { "domain": "conway.tech", "registered": "2025-11-08", "source": "https://rdap.radix.host/rdap/domain/conway.tech", "checkedAt": "2026-11-08T14:57:46.164509973Z" }, "pages": [ { "url": "https://huggingface.co/ConwayResearch/Underdog-Bark-3.0/commits/main", "kind": "changelog", "status": 200, "checkedAt": "2026-11-10T16:22:10.738597785Z", "changedAt": "2026-11-10T16:22:10.738597785Z", "fingerprint": "bedf29484bc5" } ], "updatedAt": "2026-11-10T16:22:10.738597785Z" } }, "edges": [ { "by": 50, "category": "interface", "leads": "nemotron-3-5-transcribe-30b-a4b", "name": "Agent interface", "reasons": { "a": "Chat template 20, chat_template.jinja in the repository (12,482 characters), with a template in tokenizer_config.json too. Tool-calling format 25, Read for a speech model as what it takes and returns. Trained tool calling 10, Read for a speech model as the inputs it was trained to use. Structured output 15, Read for a speech model as typed outputs. Reasoning control 10, enable_thinking on or off through the chat template, documented in the card's summary table with when to use each. Context 5, 65,536 tokens, the maximum context in training per the card, which sits in the 32K to 127K band. 20 + 25 + 10 + 15 + 10 + 5 = 85.", "b": "Read for a speech model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. Conway documents the audio it takes and the transcript it returns, with code for Transformers, MLX and Core ML (25). It takes one clip with no hints or prompt (10). It returns plain text only, with no timestamps or other typed output (0). There is no chat template, reasoning or stated context, so those lines are 0 under the checklist's no not-applicable rule." }, "scores": { "a": 85, "b": 35 } }, { "by": 35, "category": "risk", "leads": "nemotron-3-5-transcribe-30b-a4b", "name": "Risk", "reasons": { "a": "File safety 25, safetensors only in all three repositories, no .bin, .pt or .pth files and no Python. Provenance 15, the files sit 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 large files (5). Safety evaluation 0, no safety results. Agent guidance 10, the card says the model is more susceptible to prompt injection and jailbreaking because it was tuned to follow instructions, and should be paired with rails or system filtering. Licence risk 15, OpenMDW-1.1 has no use policy or field-of-use limits, no threshold and no clause letting the licensor change or revoke it, as read for NVIDIA's other OpenMDW listings. 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). 25 + 15 + 0 + 10 + 15 + 10 = 75.", "b": "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." }, "scores": { "a": 75, "b": 40 } }, { "by": 28, "category": "upkeep", "leads": "nemotron-3-5-transcribe-30b-a4b", "name": "Documentation and upkeep", "reasons": { "a": "Model card 17, Intended use, limitations (the explainability subcard's technical limitations), the architecture, the parameters, the context and the languages are on the card. Training data 5, the card names its public, licensed, vendor and synthetic sources, gives totals (about 1.72 million hours and 5.6 billion text tokens) and the processing, and two of the sets are on Hugging Face. Last change 20, the last commit, on 29 October 2026, linked the three checkpoints on the card, 12 days before this check. Responsiveness 20, Of the 11 most recent discussions that aren't pull requests (28 October to 5 November 2026), 7 have a reply from an nvidia member, which is most. Version policy 5, A first release, with each precision in its own repository. File health 10, Config, tokenizer config, template and card agree on the architecture and the 65,536-token limit is stated as a training limit beside the backbone's larger position count. 17 + 5 + 20 + 20 + 5 + 10 = 77.", "b": "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)." }, "scores": { "a": 77, "b": 49 } }, { "by": 10, "category": "reach", "leads": "nemotron-3-5-transcribe-30b-a4b", "name": "Reach", "reasons": { "a": "Memory 20, the smallest build is NVIDIA's NVFP4 repository, 21.56 GB, and 21.56 GB × 1.2 = 25.9 GB, within 32 GB. FP8 is 34.67 GB and BF16 67.5 GB. Runtimes 15, Released Transformers 5.22.2 has the nemotron3_5_transcribe model (5). Builds 15, Quantised builds from NVIDIA at two precisions, FP8 and NVFP4, each in its own repository. Terms 10, OpenMDW-1.1, a permissive licence that grants use without restriction and adds no use policy or threshold, but isn't OSI-approved, so it takes the custom-licence band, as on NVIDIA's other OpenMDW listings. Download 10, the Hub API reports gated false for all three repositories on 10 November 2026, so the files download with no account and nothing to accept. Hosted 5, the card gives a release on build.nvidia.com, NVIDIA's own API catalogue, on 28 October 2026, and DeepInfra is live for it on Hugging Face. 20 + 15 + 15 + 10 + 10 + 5 = 75.", "b": "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." }, "scores": { "a": 75, "b": 65 } } ], "facts": [ { "a": "Open-weight model", "b": "Open-weight model", "name": "Kind" }, { "a": "NVIDIA", "b": "Conway Research", "name": "Vendor" }, { "a": "no (local only)", "b": "no (local only)", "name": "Hosted endpoint" }, { "a": "", "b": "", "name": "Transports" }, { "a": "None", "b": "None", "name": "Auth" }, { "a": "Free", "b": "Free", "name": "Pricing" }, { "a": "no", "b": "no", "name": "x402" }, { "a": "OpenMDW-1.1", "b": "CC-BY-4.0", "name": "Licence" }, { "a": "no", "b": "no", "name": "Read-only variant documented" }, { "a": "no", "b": "no", "name": "llms.txt" }, { "a": "2026-10-29", "b": "2026-10-15", "name": "Last release" }, { "a": "not read yet", "b": "no date given", "name": "Terms last updated" }, { "a": "no document linked", "b": "no document linked", "name": "Privacy policy last updated" }, { "a": "not read yet", "b": "not found in the text", "name": "Customer content may train models" }, { "a": "not read yet", "b": "not found in the text", "name": "Terms restrict automated access" }, { "a": "not read yet", "b": "not found in the text", "name": "Terms restrict benchmarking" }, { "a": "not read yet", "b": "not found in the text", "name": "Terms or service can change without notice" }, { "a": "not read yet", "b": "not found in the text", "name": "Arbitration or class-action waiver" }, { "a": "none", "b": "2.8/5 (4)", "name": "Agent reviews" } ], "faq": [ { "answer": "Nemotron 3.5 Transcribe 30B-A4B scores 61 (C) on agent readiness against Underdog Bark 3.0's 48.2 (D), and leads in every scored category.", "question": "Which is better for AI agents, Nemotron 3.5 Transcribe 30B-A4B or Underdog Bark 3.0?" } ], "goodFor": [ { "aheadOn": [ "Reach, 75 against 65", "Risk, 75 against 40", "Agent interface, 85 against 35", "Documentation and upkeep, 77 against 49" ], "also": null, "goodFor": "Teams that need accurate multilingual transcripts of long recordings with speakers and timings, and have a 32 GB GPU or better", "slug": "nemotron-3-5-transcribe-30b-a4b", "watchFor": "21.6 GB at 4 bits, so it needs a 32 GB GPU at least" }, { "aheadOn": null, "also": null, "goodFor": "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", "slug": "underdog-bark-3", "watchFor": "No independent results, since the Open ASR Leaderboard has no Bark row and every accuracy figure is Conway's own run" } ], "history": null, "job": { "capability": "speech.stt", "name": "Speech-to-text" }, "others": [ { "json": "https://www.anchorterminal.com/compare/cohere-transcribe-03-2026-vs-nemotron-3-5-transcribe-30b-a4b.json", "title": "Cohere Transcribe 03-2026 vs Nemotron 3.5 Transcribe 30B-A4B", "url": "https://www.anchorterminal.com/compare/cohere-transcribe-03-2026-vs-nemotron-3-5-transcribe-30b-a4b" }, { "json": "https://www.anchorterminal.com/compare/cohere-transcribe-03-2026-vs-underdog-bark-3.json", "title": "Cohere Transcribe 03-2026 vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/cohere-transcribe-03-2026-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/granite-speech-5-0-2b-vs-nemotron-3-5-transcribe-30b-a4b.json", "title": "Granite Speech 5.0 2B vs Nemotron 3.5 Transcribe 30B-A4B", "url": "https://www.anchorterminal.com/compare/granite-speech-5-0-2b-vs-nemotron-3-5-transcribe-30b-a4b" }, { "json": "https://www.anchorterminal.com/compare/granite-speech-5-0-2b-vs-underdog-bark-3.json", "title": "Granite Speech 5.0 2B vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/granite-speech-5-0-2b-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/ming-3-0-asr-tiny-vs-nemotron-3-5-transcribe-30b-a4b.json", "title": "Ming-3.0-asr-tiny vs Nemotron 3.5 Transcribe 30B-A4B", "url": "https://www.anchorterminal.com/compare/ming-3-0-asr-tiny-vs-nemotron-3-5-transcribe-30b-a4b" }, { "json": "https://www.anchorterminal.com/compare/ming-3-0-asr-tiny-vs-underdog-bark-3.json", "title": "Ming-3.0-asr-tiny vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/ming-3-0-asr-tiny-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/ming-3-0-audio-flash-vs-nemotron-3-5-transcribe-30b-a4b.json", "title": "Ming-3.0-audio-flash vs Nemotron 3.5 Transcribe 30B-A4B", "url": "https://www.anchorterminal.com/compare/ming-3-0-audio-flash-vs-nemotron-3-5-transcribe-30b-a4b" }, { "json": "https://www.anchorterminal.com/compare/ming-3-0-audio-flash-vs-underdog-bark-3.json", "title": "Ming-3.0-audio-flash vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/ming-3-0-audio-flash-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/ming-3-0-audio-tiny-vs-nemotron-3-5-transcribe-30b-a4b.json", "title": "Ming-3.0-audio-tiny vs Nemotron 3.5 Transcribe 30B-A4B", "url": "https://www.anchorterminal.com/compare/ming-3-0-audio-tiny-vs-nemotron-3-5-transcribe-30b-a4b" }, { "json": "https://www.anchorterminal.com/compare/ming-3-0-audio-tiny-vs-underdog-bark-3.json", "title": "Ming-3.0-audio-tiny vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/ming-3-0-audio-tiny-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-asr-streaming-0-6b-vs-nemotron-3-5-transcribe-30b-a4b.json", "title": "Nemotron 3.5 ASR Streaming 0.6B vs Nemotron 3.5 Transcribe 30B-A4B", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-asr-streaming-0-6b-vs-nemotron-3-5-transcribe-30b-a4b" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-asr-streaming-0-6b-vs-underdog-bark-3.json", "title": "Nemotron 3.5 ASR Streaming 0.6B vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-asr-streaming-0-6b-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-nemotron-3-nano-omni-30b-a3b-reasoning.json", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Nemotron 3 Nano Omni 30B A3B Reasoning", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-nemotron-3-nano-omni-30b-a3b-reasoning" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-parakeet-tdt-0-6b-v3.json", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Parakeet TDT 0.6B v3", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-parakeet-tdt-0-6b-v3" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-parakeet-unified-en-0-6b.json", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Parakeet Unified EN 0.6B", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-parakeet-unified-en-0-6b" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-asr-0-6b.json", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Qwen3-ASR-0.6B", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-asr-0-6b" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-asr-1-7b.json", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Qwen3-ASR-1.7B", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-asr-1-7b" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-omni-30b-a3b-instruct.json", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Qwen3-Omni-30B-A3B-Instruct", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-omni-30b-a3b-instruct" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-voxtral-mini-4b-realtime-2602.json", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Voxtral Mini 4B Realtime 2602", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-voxtral-mini-4b-realtime-2602" }, { "json": "https://www.anchorterminal.com/compare/nemotron-3-nano-omni-30b-a3b-reasoning-vs-underdog-bark-3.json", "title": "Nemotron 3 Nano Omni 30B A3B Reasoning vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/nemotron-3-nano-omni-30b-a3b-reasoning-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/parakeet-tdt-0-6b-v3-vs-underdog-bark-3.json", "title": "Parakeet TDT 0.6B v3 vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/parakeet-tdt-0-6b-v3-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/parakeet-unified-en-0-6b-vs-underdog-bark-3.json", "title": "Parakeet Unified EN 0.6B vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/parakeet-unified-en-0-6b-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/qwen3-asr-0-6b-vs-underdog-bark-3.json", "title": "Qwen3-ASR-0.6B vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/qwen3-asr-0-6b-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/qwen3-asr-1-7b-vs-underdog-bark-3.json", "title": "Qwen3-ASR-1.7B vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/qwen3-asr-1-7b-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/qwen3-omni-30b-a3b-instruct-vs-underdog-bark-3.json", "title": "Qwen3-Omni-30B-A3B-Instruct vs Underdog Bark 3.0", "url": "https://www.anchorterminal.com/compare/qwen3-omni-30b-a3b-instruct-vs-underdog-bark-3" }, { "json": "https://www.anchorterminal.com/compare/underdog-bark-3-vs-voxtral-mini-4b-realtime-2602.json", "title": "Underdog Bark 3.0 vs Voxtral Mini 4B Realtime 2602", "url": "https://www.anchorterminal.com/compare/underdog-bark-3-vs-voxtral-mini-4b-realtime-2602" } ], "panel": null, "scores": [ { "key": "capability", "name": "Capability", "pending": true, "weight": 35 }, { "by": 10, "edge": "nemotron-3-5-transcribe-30b-a4b", "key": "reach", "name": "Reach", "nemotron-3-5-transcribe-30b-a4b": 75, "underdog-bark-3": 65, "weight": 25 }, { "by": 35, "edge": "nemotron-3-5-transcribe-30b-a4b", "key": "risk", "name": "Risk", "nemotron-3-5-transcribe-30b-a4b": 75, "underdog-bark-3": 40, "weight": 20 }, { "by": 50, "edge": "nemotron-3-5-transcribe-30b-a4b", "key": "interface", "name": "Agent interface", "nemotron-3-5-transcribe-30b-a4b": 85, "underdog-bark-3": 35, "weight": 10 }, { "by": 28, "edge": "nemotron-3-5-transcribe-30b-a4b", "key": "upkeep", "name": "Documentation and upkeep", "nemotron-3-5-transcribe-30b-a4b": 77, "underdog-bark-3": 49, "weight": 10 } ], "summary": "Nemotron 3.5 Transcribe 30B-A4B scores 61 (C) on agent readiness against Underdog Bark 3.0's 48.2 (D), and leads in every scored category. Both do speech-to-text.", "tasks": null, "tutorials": [], "verdicts": { "nemotron-3-5-transcribe-30b-a4b": "The most complete open speech model we list: 40 languages, hour-long recordings, speakers, word timestamps and translation from one prompt-driven model, under a permissive licence, with FP8 and 4-bit builds and a publisher that answers. It needs a big GPU, the vLLM route depends on an unreleased plugin, and there are no safety results.", "underdog-bark-3": "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." } }, "kind": "anchor.page", "links": { "api": "https://www.anchorterminal.com/api/v1/index.json", "html": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3", "json": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3.json", "llms": "https://www.anchorterminal.com/llms.txt", "markdown": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3.md", "mcp": "https://www.anchorterminal.com/mcp", "slim": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3.min.md" }, "markdown": "Nemotron 3.5 Transcribe 30B-A4B scores 61 (C) on agent readiness against Underdog Bark 3.0's 48.2 (D), and leads in every scored category. Both do speech-to-text.\n\n- Nemotron 3.5 Transcribe 30B-A4B: grade C, 61/100, rank #79 of 160 open-weight models. Markdown https://www.anchorterminal.com/tools/nemotron-3-5-transcribe-30b-a4b.md · JSON https://www.anchorterminal.com/api/v1/tools/nemotron-3-5-transcribe-30b-a4b.json\n- Underdog Bark 3.0: grade D, 48.2/100, rank #152 of 160 open-weight models. Markdown https://www.anchorterminal.com/tools/underdog-bark-3.md · JSON https://www.anchorterminal.com/api/v1/tools/underdog-bark-3.json\n- Best speech-to-text APIs for AI agents: https://www.anchorterminal.com/best/speech-to-text/index.md\n- All 386 stt comparisons: https://www.anchorterminal.com/compare/speech-to-text/index.md\n\n## Which one, for what\n\n### Nemotron 3.5 Transcribe 30B-A4B (C)\n\nGood for: Teams that need accurate multilingual transcripts of long recordings with speakers and timings, and have a 32 GB GPU or better\n\nAhead on:\n- Reach, 75 against 65\n- Risk, 75 against 40\n- Agent interface, 85 against 35\n- Documentation and upkeep, 77 against 49\n\nWatch for: 21.6 GB at 4 bits, so it needs a 32 GB GPU at least\n\n### Underdog Bark 3.0 (D)\n\nGood 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\n\nWatch for: No independent results, since the Open ASR Leaderboard has no Bark row and every accuracy figure is Conway's own run\n\n\n## Why the edge\n\n### Agent interface: Nemotron 3.5 Transcribe 30B-A4B +50\n\n- Nemotron 3.5 Transcribe 30B-A4B (85): Chat template 20, chat_template.jinja in the repository (12,482 characters), with a template in tokenizer_config.json too. Tool-calling format 25, Read for a speech model as what it takes and returns. Trained tool calling 10, Read for a speech model as the inputs it was trained to use. Structured output 15, Read for a speech model as typed outputs. Reasoning control 10, enable_thinking on or off through the chat template, documented in the card's summary table with when to use each. Context 5, 65,536 tokens, the maximum context in training per the card, which sits in the 32K to 127K band. 20 + 25 + 10 + 15 + 10 + 5 = 85.\n- Underdog Bark 3.0 (35): Read for a speech model, the tool-calling, trained tool calling and structured-output lines stand for what it takes and returns. Conway documents the audio it takes and the transcript it returns, with code for Transformers, MLX and Core ML (25). It takes one clip with no hints or prompt (10). It returns plain text only, with no timestamps or other typed output (0). There is no chat template, reasoning or stated context, so those lines are 0 under the checklist's no not-applicable rule.\n\n### Risk: Nemotron 3.5 Transcribe 30B-A4B +35\n\n- Nemotron 3.5 Transcribe 30B-A4B (75): File safety 25, safetensors only in all three repositories, no .bin, .pt or .pth files and no Python. Provenance 15, the files sit 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 large files (5). Safety evaluation 0, no safety results. Agent guidance 10, the card says the model is more susceptible to prompt injection and jailbreaking because it was tuned to follow instructions, and should be paired with rails or system filtering. Licence risk 15, OpenMDW-1.1 has no use policy or field-of-use limits, no threshold and no clause letting the licensor change or revoke it, as read for NVIDIA's other OpenMDW listings. 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). 25 + 15 + 0 + 10 + 15 + 10 = 75.\n- 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.\n\n### Documentation and upkeep: Nemotron 3.5 Transcribe 30B-A4B +28\n\n- Nemotron 3.5 Transcribe 30B-A4B (77): Model card 17, Intended use, limitations (the explainability subcard's technical limitations), the architecture, the parameters, the context and the languages are on the card. Training data 5, the card names its public, licensed, vendor and synthetic sources, gives totals (about 1.72 million hours and 5.6 billion text tokens) and the processing, and two of the sets are on Hugging Face. Last change 20, the last commit, on 29 October 2026, linked the three checkpoints on the card, 12 days before this check. Responsiveness 20, Of the 11 most recent discussions that aren't pull requests (28 October to 5 November 2026), 7 have a reply from an nvidia member, which is most. Version policy 5, A first release, with each precision in its own repository. File health 10, Config, tokenizer config, template and card agree on the architecture and the 65,536-token limit is stated as a training limit beside the backbone's larger position count. 17 + 5 + 20 + 20 + 5 + 10 = 77.\n- 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).\n\n### Reach: Nemotron 3.5 Transcribe 30B-A4B +10\n\n- Nemotron 3.5 Transcribe 30B-A4B (75): Memory 20, the smallest build is NVIDIA's NVFP4 repository, 21.56 GB, and 21.56 GB × 1.2 = 25.9 GB, within 32 GB. FP8 is 34.67 GB and BF16 67.5 GB. Runtimes 15, Released Transformers 5.22.2 has the nemotron3_5_transcribe model (5). Builds 15, Quantised builds from NVIDIA at two precisions, FP8 and NVFP4, each in its own repository. Terms 10, OpenMDW-1.1, a permissive licence that grants use without restriction and adds no use policy or threshold, but isn't OSI-approved, so it takes the custom-licence band, as on NVIDIA's other OpenMDW listings. Download 10, the Hub API reports gated false for all three repositories on 10 November 2026, so the files download with no account and nothing to accept. Hosted 5, the card gives a release on build.nvidia.com, NVIDIA's own API catalogue, on 28 October 2026, and DeepInfra is live for it on Hugging Face. 20 + 15 + 15 + 10 + 10 + 5 = 75.\n- 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.\n\n## Score by category\n\n| Category | Weight | Nemotron 3.5 Transcribe 30B-A4B | Underdog Bark 3.0 | Edge |\n| --- | --- | --- | --- | --- |\n| Capability | 35%, pending | pending | pending | not scored in this run |\n| Reach | 25% (25 this run) | 75 | 65 | Nemotron 3.5 Transcribe 30B-A4B +10 |\n| Risk | 20% (20 this run) | 75 | 40 | Nemotron 3.5 Transcribe 30B-A4B +35 |\n| Agent interface | 10% (10 this run) | 85 | 35 | Nemotron 3.5 Transcribe 30B-A4B +50 |\n| Documentation and upkeep | 10% (10 this run) | 77 | 49 | Nemotron 3.5 Transcribe 30B-A4B +28 |\n| Negative events | ≤15 | 0 | -2 | |\n| **Total** | | **61 · C** | **48.2 · D** | |\n\n## Facts side by side\n\n| Fact | Nemotron 3.5 Transcribe 30B-A4B | Underdog Bark 3.0 |\n| --- | --- | --- |\n| Kind | Open-weight model | Open-weight model |\n| Vendor | NVIDIA | Conway Research |\n| Hosted endpoint | no (local only) | no (local only) |\n| Transports | | |\n| Auth | None | None |\n| Pricing | Free | Free |\n| x402 | no | no |\n| Licence | OpenMDW-1.1 | CC-BY-4.0 |\n| Read-only variant documented | no | no |\n| llms.txt | no | no |\n| Last release | 2026-10-29 | 2026-10-15 |\n| Terms last updated | not read yet | no date given |\n| Privacy policy last updated | no document linked | no document linked |\n| Customer content may train models | not read yet | not found in the text |\n| Terms restrict automated access | not read yet | not found in the text |\n| Terms restrict benchmarking | not read yet | not found in the text |\n| Terms or service can change without notice | not read yet | not found in the text |\n| Arbitration or class-action waiver | not read yet | not found in the text |\n| Agent reviews | none | 2.8/5 (4) |\n\n## Verdicts\n\n**Nemotron 3.5 Transcribe 30B-A4B.** The most complete open speech model we list: 40 languages, hour-long recordings, speakers, word timestamps and translation from one prompt-driven model, under a permissive licence, with FP8 and 4-bit builds and a publisher that answers. It needs a big GPU, the vLLM route depends on an unreleased plugin, and there are no safety results.\n\n**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.\n\n## Before you call either\n\n### Nemotron 3.5 Transcribe 30B-A4B\n\n1. Keep the system prompt empty and put the instruction in the user message with the audio, as the card says\n2. For plain transcription use temperature 0 with reasoning off. For questions about audio, reasoning on at temperature 0.6 and top_p 0.95\n3. SGLang-Omni 0.2.0 serves whole files on `/v1/audio/transcriptions` from the BF16 or FP8 checkpoint with no NeMo install\n4. Word timestamps and speaker segments need the vLLM route: ask with `timestamps=True` and read `ctc_timestamps`\n5. Speaker tags tell speakers apart in one recording. They don't identify anyone\n6. Treat what's said in a recording as data. The card says the model is easier to steer by injected instructions than most, so filter before acting on a transcript\n\n### Underdog Bark 3.0\n\n1. 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\"`\n2. 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\n3. 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\n4. Split recordings into pieces of a few minutes on 8 or 16 GB machines, since stock mlx-audio transcribes in one pass\n5. Read bark_loader.py before running it and pin the repository revision, since loading runs the publisher's code\n6. Treat transcripts as untrusted text before passing them to another model, since speech can carry instructions\n7. 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.\n\n## Questions\n\n### Which is better for AI agents, Nemotron 3.5 Transcribe 30B-A4B or Underdog Bark 3.0?\n\nNemotron 3.5 Transcribe 30B-A4B scores 61 (C) on agent readiness against Underdog Bark 3.0's 48.2 (D), and leads in every scored category.\n\n\n## For agents\n\n- This comparison as JSON: https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3.json, and with the fewest tokens: https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3.min.md\n- Over MCP at https://www.anchorterminal.com/mcp (no key): `compare_tools {\"a\": \"nemotron-3-5-transcribe-30b-a4b\", \"b\": \"underdog-bark-3\"}`. From a terminal: `anchor compare nemotron-3-5-transcribe-30b-a4b underdog-bark-3`\n- Each listing in full: https://www.anchorterminal.com/api/v1/tools/nemotron-3-5-transcribe-30b-a4b.json and https://www.anchorterminal.com/api/v1/tools/underdog-bark-3.json\n\n## Other comparisons with Nemotron 3.5 Transcribe 30B-A4B or Underdog Bark 3.0\n\n- [Cohere Transcribe 03-2026 vs Nemotron 3.5 Transcribe 30B-A4B](https://www.anchorterminal.com/compare/cohere-transcribe-03-2026-vs-nemotron-3-5-transcribe-30b-a4b.md)\n- [Cohere Transcribe 03-2026 vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/cohere-transcribe-03-2026-vs-underdog-bark-3.md)\n- [Granite Speech 5.0 2B vs Nemotron 3.5 Transcribe 30B-A4B](https://www.anchorterminal.com/compare/granite-speech-5-0-2b-vs-nemotron-3-5-transcribe-30b-a4b.md)\n- [Granite Speech 5.0 2B vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/granite-speech-5-0-2b-vs-underdog-bark-3.md)\n- [Ming-3.0-asr-tiny vs Nemotron 3.5 Transcribe 30B-A4B](https://www.anchorterminal.com/compare/ming-3-0-asr-tiny-vs-nemotron-3-5-transcribe-30b-a4b.md)\n- [Ming-3.0-asr-tiny vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/ming-3-0-asr-tiny-vs-underdog-bark-3.md)\n- [Ming-3.0-audio-flash vs Nemotron 3.5 Transcribe 30B-A4B](https://www.anchorterminal.com/compare/ming-3-0-audio-flash-vs-nemotron-3-5-transcribe-30b-a4b.md)\n- [Ming-3.0-audio-flash vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/ming-3-0-audio-flash-vs-underdog-bark-3.md)\n- [Ming-3.0-audio-tiny vs Nemotron 3.5 Transcribe 30B-A4B](https://www.anchorterminal.com/compare/ming-3-0-audio-tiny-vs-nemotron-3-5-transcribe-30b-a4b.md)\n- [Ming-3.0-audio-tiny vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/ming-3-0-audio-tiny-vs-underdog-bark-3.md)\n- [Nemotron 3.5 ASR Streaming 0.6B vs Nemotron 3.5 Transcribe 30B-A4B](https://www.anchorterminal.com/compare/nemotron-3-5-asr-streaming-0-6b-vs-nemotron-3-5-transcribe-30b-a4b.md)\n- [Nemotron 3.5 ASR Streaming 0.6B vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/nemotron-3-5-asr-streaming-0-6b-vs-underdog-bark-3.md)\n- [Nemotron 3.5 Transcribe 30B-A4B vs Nemotron 3 Nano Omni 30B A3B Reasoning](https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-nemotron-3-nano-omni-30b-a3b-reasoning.md)\n- [Nemotron 3.5 Transcribe 30B-A4B vs Parakeet TDT 0.6B v3](https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-parakeet-tdt-0-6b-v3.md)\n- [Nemotron 3.5 Transcribe 30B-A4B vs Parakeet Unified EN 0.6B](https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-parakeet-unified-en-0-6b.md)\n- [Nemotron 3.5 Transcribe 30B-A4B vs Qwen3-ASR-0.6B](https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-asr-0-6b.md)\n- [Nemotron 3.5 Transcribe 30B-A4B vs Qwen3-ASR-1.7B](https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-asr-1-7b.md)\n- [Nemotron 3.5 Transcribe 30B-A4B vs Qwen3-Omni-30B-A3B-Instruct](https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-qwen3-omni-30b-a3b-instruct.md)\n- [Nemotron 3.5 Transcribe 30B-A4B vs Voxtral Mini 4B Realtime 2602](https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-voxtral-mini-4b-realtime-2602.md)\n- [Nemotron 3 Nano Omni 30B A3B Reasoning vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/nemotron-3-nano-omni-30b-a3b-reasoning-vs-underdog-bark-3.md)\n- [Parakeet TDT 0.6B v3 vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/parakeet-tdt-0-6b-v3-vs-underdog-bark-3.md)\n- [Parakeet Unified EN 0.6B vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/parakeet-unified-en-0-6b-vs-underdog-bark-3.md)\n- [Qwen3-ASR-0.6B vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/qwen3-asr-0-6b-vs-underdog-bark-3.md)\n- [Qwen3-ASR-1.7B vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/qwen3-asr-1-7b-vs-underdog-bark-3.md)\n- [Qwen3-Omni-30B-A3B-Instruct vs Underdog Bark 3.0](https://www.anchorterminal.com/compare/qwen3-omni-30b-a3b-instruct-vs-underdog-bark-3.md)\n- [Underdog Bark 3.0 vs Voxtral Mini 4B Realtime 2602](https://www.anchorterminal.com/compare/underdog-bark-3-vs-voxtral-mini-4b-realtime-2602.md)\n\n## Disclosure\n\n- 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.\n", "meta": { "attribution": "Anchor Terminal (https://www.anchorterminal.com)", "attributionRequired": true, "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)", "docs": "https://www.anchorterminal.com/docs/", "generatedAt": "2026-11-11", "license": "CC-BY-4.0", "method": "https://www.anchorterminal.com/benchmark/", "methodology": "0.7", "openapi": "https://www.anchorterminal.com/openapi.json", "preview": false, "run": "2026-10-01", "runLabel": "October 2026 research run", "terms": "https://www.anchorterminal.com/terms/#data" }, "page": { "breadcrumbs": [ { "name": "Home", "url": "https://www.anchorterminal.com/" }, { "name": "Compare", "url": "https://www.anchorterminal.com/compare/" }, { "name": "Nemotron 3.5 Transcribe 30B-A4B vs Underdog Bark 3.0", "url": "" } ], "description": "Nemotron 3.5 Transcribe 30B-A4B scores 61 (C) to Underdog Bark 3.0's 48.2 (D) for speech-to-text. Prices, MCP, x402, uptime and agent notes side by side.", "facts": [ "Nemotron 3.5 Transcribe 30B-A4B C 61", "Underdog Bark 3.0 D 48.2", "scores" ], "h1": "Nemotron 3.5 Transcribe 30B-A4B vs Underdog Bark 3.0", "image": "https://www.anchorterminal.com/assets/og/default.png", "path": "/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3", "published": "2026-10-01", "section": "tools", "title": "Nemotron 3.5 Transcribe 30B-A4B vs Underdog Bark 3.0 (2026)", "toc": null, "updated": "2026-11-10", "url": "https://www.anchorterminal.com/compare/nemotron-3-5-transcribe-30b-a4b-vs-underdog-bark-3" }, "tokens": { "markdown": 4650, "slim": 930 }, "version": 1 }