Argmax Pro SDK 3

The next generation of our flagship SDK is here with support for the latest frontier speaker and speech recognition models!

• Real-time speaker recognition with up to 8 speakers with Nemotron 3 Diarization
• Multilingual speech-to-text across 30 languages with Qwen3-ASR

‍

‍

Frontier Speaker Recognition in Real-time with 8 Speakers

NVIDIA is launching Nemotron 3 Diarization today. Argmax is thrilled to be a launch partner and add Day 0 support in Argmax SDK! This model is the successor to Sortformer v2 with significantly upgraded capabilities:

• Maximum speaker limit is raised from 4 to 8, covering most use cases.
• Speaker consistency holds even under challenging conditions such as ambient recordings with far-field speakers, background noise and muffled speech

‍

Reproducible Benchmarks

We have been testing this model for several weeks on both private real-world datasets as well as public benchmarks on OpenBench and it consistently ranked top by a large margin. Many Argmax customers who were previously using Sortformer v2 are already upgrading to Nemotron 3 Diarization.

In specific:

• We benchmarked 6 speaker diarization systems across 11 datasets representing diverse and challenging real-world conditions using OpenBench.
• Nemotron 3 Diarization achieved the lowest error rate (DER) across all 6 systems.
• It also scored the lowest across 5 of the 11 datasets despite its maximum speaker constraint of 8.
• This represents a huge leap in accuracy compared to its predecessor Sortformer v2.1 whose error rate was 60% higher.

‍

Pre-diarized Transcription API

Before Nemotron 3 Diarization, Argmax SDK has assigned speakers to transcribed words after the fact to avoid compounding speaker and speech recognition errors.

For the first time, applying speaker diarization with Nemotron before transcription leads to better transcripts and we have built the new Pre-diarized Transcription API exclusively for Nemotron. The demo video above uses this new API.

‍

Frontier Speech-to-Text Across 30 Languages

Qwen3-ASR achieves market-leading speech-to-text accuracy across 30 languages. Argmax SDK implementation comes with word timestamps, custom vocabulary, code-switching and real-time API support, all in one model!

The best part: This model works with the existing Argmax SDK APIs to facilitate seamless upgrades.

NVIDIA Parakeet remains the fastest model in Argmax SDK with the second highest accuracy after Qwen3-ASR.

‍

Getting Started

Test these models on your own by installing Argmax Playground on your Mac or iPhone here.

If you are attending Interspeech 2026 in Sydney on September 27, come talk to the Argmax team at the Expo and get live demos of frontier voice AI apps built with Argmax SDK 3!