· Hardware

B200

NVIDIA Blackwell · validated on NVIDIA B200 · 12 models in production · last verified Aug 31, 2026.

ModelPrecisionCapabilitiesSinceLast verifiedBenchmark
Gemma 4 E2BGemmaBF16Query · Captionv2.1Aug 30, 2026Results
Gemma 4 E4BGemmaBF16Query · Captionv2.1Aug 30, 2026Results
Moondream 2Moondream VLMBF16Query · Caption · Detect · Point · Segmentv2.1Aug 31, 2026
Moondream 3Moondream VLMBF16Query · Caption · Detect · Point · Segmentv2.1Aug 30, 2026Results
Parakeet TDT 0.6B v3ParakeetBF16Transcribev2.1Aug 31, 2026Results
Qwen 3.5 0.8BQwen 3BF16Query · Captionv2.1Aug 30, 2026Results
Qwen 3.5 2BQwen 3BF16Query · Captionv2.1Aug 30, 2026Results
Qwen 3.5 4BQwen 3BF16Query · Captionv2.1Aug 30, 2026Results
Qwen 3.5 9BQwen 3BF16Query · Captionv2.1Aug 30, 2026Results
Qwen3-ASR 0.6BQwen3-ASRBF16Transcribev2.1Aug 31, 2026Results
Qwen3-ASR 1.7BQwen3-ASRBF16Transcribev2.1Aug 31, 2026Results
Whisper large-v3-turboWhisperBF16Transcribev2.1Aug 31, 2026Results
Run on B200

Install the SDK and run any listed model locally.

$pip install moondream

Quickstart docs · all architectures · product updates