Side by side

See the difference.

Scores, capabilities and costs for the models you’re considering.

Scroll to compare models →

Comparison of tts-1, Moondream: Moondream 3 Preview, abr-ai/niagara-38m-batch.en
3 modelsUnknowntts-1Remove ×moondreamMoondream: Moondream 3 PreviewRemove ×abr-aiabr-ai/niagara-38m-batch.enRemove ×
What’s differentMoondream 3 Preview is a vision-language model from Moondream with 9B total parameters and 2B active (mixture-of-experts). It handles visual question answering, captioning, OCR, and visual reasoning over…0.038B parameters · 1 language · Encoder: State-space model · Decoder: CTC
DateOct 16, 2026DiscoveredOct 15, 2026ListedOct 17, 2026Discovered
Listed rateSee provider quotes$0.30 in / $2.50 out/ 1M tokensPrimeIntellect · via OpenRouter · all routes ↗primeintellect/fp8 · quantization=fp8See provider quotes
Provider endpoints0 endpoints →2 endpoints →0 endpoints →
AMI-Cleaned WER ↗English short-form; normalization unversionedNot reportedNot reported13.87% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
Earnings22-Cleaned-AA-chunked WER ↗English short-form; normalization unversionedNot reportedNot reported12.94% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
Gigaspeech-Cleaned WER ↗English short-form; normalization unversionedNot reportedNot reported11.38% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
LS Clean WER ↗English short-form; normalization unversionedNot reportedNot reported3.81% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
LS Other WER ↗English short-form; normalization unversionedNot reportedNot reported9.28% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
SPGISpeech WER ↗English short-form; normalization unversionedNot reportedNot reported3.08% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
URGENT2024 WER ↗English short-form; normalization unversionedNot reportedNot reported22.88% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
Voice Arena Monsoon WER ↗English short-form; normalization unversionedNot reportedNot reported10.41% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
Voxpopuli-AA-Cleaned WER ↗English short-form; normalization unversionedNot reportedNot reported5.16% WER ↗abr-ai/niagara-38m-batch.enLeaderboard entry
Context window—32K—
Max output—4.1K—
Parameters——0.038B
Image inputNot reportedSupportedNot reported
Audio inputNot reportedNot reportedSupported
TranscriptionNot reportedNot reportedSupported
ReasoningNot reportedSupportedNot reported
Open weightsNot reportedSupportedNot reported

Add another model

Search all models →