# Apriel-Voice-27B-preview

> An open model for speech, reasoning, and tool use from ServiceNow

- **Organisation:** ServiceNow-AI
- **Category:** Omni / Any-to-Any
- **Parameters:** 27.2B
- **Context length:** 262,144 tokens
- **Licence:** MIT (Commercial use OK)
- **Released:** 2026-10-13
- **Hugging Face:** 1.5K downloads (30 days), 97 likes

## Estimated memory

| Quant | ≈ GB |
|---|---|
| Q2_K | 11.13 |
| Q3_K_M | 15.39 |
| Q4_K_M | 18.52 |
| Q5_K_M | 21.31 |
| Q6_K | 24.31 |
| Q8_0 | 30.67 |
| F16 | 55.63 |

## Runs with

- [Transformers](https://github.com/huggingface/transformers)

## About

An open model for speech, reasoning, and tool use from ServiceNow

Apriel Voice 27B Preview understands spoken requests, reasons over conversational context, and generates text responses and tool calls through a shared Gemma 4 backbone. It adds a speech interface while retaining text and image input capabilities. Pair it with an external speech synthesizer to build interactive voice assistants, customer support workflows, and spoken tool use.

Try Apriel Live · Read the blog · Technical report

The model supports transcription, speech translation, spoken instruction following, reasoning, and function calling. The application executes tools and returns their results to the conversation. Real-time voice interaction is supported by the application and serving stack; the checkpoint itself generates text and tool calls rather than native full-duplex speech.

A Whisper large-v3-turbo speech encoder connects to the Gemma 4 26B-A4B backbone through a learned projector. Training first aligns this speech interface, then adapts it with the backbone using speech tasks and text replay. Cross-entropy and teacher-distillation supervision vary by data source; the technical report describes the full recipe.

## Alternatives

- [nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16](https://selfhost.directory/models/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning-bf16) (33B (3B active), nvidia-open-model-agreement, ≈20.48 GB at Q4): NVIDIA Nemotron 3 Nano Omni is a multimodal large language model that unifies video, audio, image, and text understanding to support enterprise-grade Q&A, summarization, transcription, and document intelligence workflows.
- [Qwen/Qwen3-Omni-30B-A3B-Instruct](https://selfhost.directory/models/qwen/qwen3-omni-30b-a3b-instruct) (35.3B (3B active), Apache 2.0, ≈22.15 GB at Q4): Qwen3-Omni is the natively end-to-end multilingual omni-modal foundation models. It processes text, images, audio, and video, and delivers real-time streaming responses in both text and natural speech. We introduce several architectural upgrades to improve performance and efficiency. Key features:
- [Qwen/Qwen3-Omni-30B-A3B-Thinking](https://selfhost.directory/models/qwen/qwen3-omni-30b-a3b-thinking) (31.7B (3B active), Apache 2.0, ≈20.05 GB at Q4): Qwen3-Omni is the natively end-to-end multilingual omni-modal foundation models. It processes text, images, audio, and video, and delivers real-time streaming responses in both text and natural speech. We introduce several architectural upgrades to improve performance and efficiency. Key features:
- [google/gemma-4-12B-it](https://selfhost.directory/models/google/gemma-4-12b-it) (12B, Apache 2.0, ≈10.59 GB at Q4): Gemma is a family of open models built by Google DeepMind. Gemma 4 models are multimodal, handling text and image input (with audio supported on E2B, E4B, and 12B) and generating text output. This release includes open-weights models in both pre-trained and instruction-tuned variants.
- [nvidia/NVIDIA-NemotronLabs-AI-for-Media-Sports-Tennis](https://selfhost.directory/models/nvidia/nvidia-nemotronlabs-ai-for-media-sports-tennis) (33B, openmdw-1.1, ≈20.48 GB at Q4): NVIDIA-NemotronLabs-AI-for-Media-Sports-Tennis is a fine-tuned multimodal large language model derived from Nemotron 3 Nano Omni. It was further trained on an in-house NVIDIA tennis dataset of manually labeled point-level video clips to specialize in tennis match understanding.
- [nvidia/NVIDIA-NemotronLabs-AI-for-Media-Sports-Basketball](https://selfhost.directory/models/nvidia/nvidia-nemotronlabs-ai-for-media-sports-basketball) (33B, openmdw-1.1, ≈20.48 GB at Q4): NVIDIA-NemotronLabs-AI-for-Media-Sports-Basketball is a fine-tuned multimodal large language model derived from Nemotron 3 Nano Omni.

More: https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview/alternatives

## Runs well on

- [Tesla P40](https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview/on/tesla-p40): Q5_K_M, est. 9.9 tok/s
- [Radeon RX 7900 XTX](https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview/on/rx-7900-xtx): Q5_K_M, est. 27 tok/s
- [Instinct MI50 32GB](https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview/on/instinct-mi50): Q6_K, est. 25 tok/s
- [GeForce RTX 5090](https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview/on/rtx-5090): Q6_K, est. 44 tok/s
- [Apple M4 Pro (48 GB)](https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview/on/apple-m4-pro-48gb): Q8_0, est. 5.2 tok/s
- [RTX A6000](https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview/on/rtx-a6000): Q8_0, est. 15 tok/s

## Similar size

- [Qwen/Qwen3-Omni-30B-A3B-Instruct](https://selfhost.directory/models/qwen/qwen3-omni-30b-a3b-instruct) (35.3B)
- [nvidia/Nemotron-3-Nano-Omni-30B-A3B-Reasoning-BF16](https://selfhost.directory/models/nvidia/nemotron-3-nano-omni-30b-a3b-reasoning-bf16) (33B)
- [Qwen/Qwen3-Omni-30B-A3B-Thinking](https://selfhost.directory/models/qwen/qwen3-omni-30b-a3b-thinking) (31.7B)
- [mistralai/Voxtral-Small-24B-2507](https://selfhost.directory/models/mistralai/voxtral-small-24b-2507) (24.3B)
- [Qwen/Qwen3-Omni-30B-A3B-Captioner](https://selfhost.directory/models/qwen/qwen3-omni-30b-a3b-captioner) (31.7B)
- [stepfun-ai/Step-Audio-R1](https://selfhost.directory/models/stepfun-ai/step-audio-r1) (33.5B)

## FAQ

**How much VRAM does Apriel-Voice-27B-preview need?**
About 19 GB at Q4_K_M with an 8K context. Q8_0 needs about 31 GB and full precision about 56 GB. Longer contexts need more memory for the KV cache.

**Can Apriel-Voice-27B-preview run on the Tesla P40?**
Yes. It fits at Q5_K_M (about 21 GB), at an estimated 9.9 tokens/s. It fits comfortably on the Radeon RX 7900 XTX, Instinct MI50 32GB, RTX 5090.

**Can I use Apriel-Voice-27B-preview commercially?**
Yes. It is released under MIT. Permissive: do almost anything as long as you keep the copyright notice.

**What is the context length of Apriel-Voice-27B-preview?**
262,144 tokens (256K). The full window needs far more memory than an 8K context.

## Links

- huggingface: https://huggingface.co/ServiceNow-AI/Apriel-Voice-27B-preview
- homepage: https://servicenow-ai-apriel.hf.space/blog/apriel-voice-preview

---
Source: https://selfhost.directory/models/servicenow-ai/apriel-voice-27b-preview