Model library / pipecat-ai
PhoneLLM Alpha 1 is an open-weights MoE language model fine-tuned for low-latency voice agent applications.
6 of 128 experts active per token
Context length in tokens, as published
Weights at the released precision
Licence declared on the repository
Overview
PhoneLLM Alpha 1 is a 31.6 billion-parameter mixture-of-experts language model based on Nemotron 3 Nano, released by pipecat-ai. The publisher describes it as an open-weights model fine-tuned for low-latency, multi-turn voice agent workloads such as customer-service calls in finance, healthcare, retail, hospitality and insurance sectors overall.
The model contains 31.6 B parameters, stored in bfloat16, and uses 128 experts with six active per token. It supports a 262 k token context window and is licensed under BSD-2-Clause, allowing unrestricted commercial use. Deployment requires hardware capable of MoE inference; the official NVFP4 checkpoint targets NVIDIA Blackwell GPUs with BF16 KV cache.
Specification
| Repository | pipecat-ai/phonellm-alpha-1 |
| Publisher | pipecat-ai |
| Published | 24 August 2026 |
| Architecture | NemotronHForCausalLM |
| Model type | nemotron_h |
| Parameters | 31,577,937,344 |
| Experts | 128 total, 6 active per token |
| Layers | 52 |
| Hidden size | 2,688 |
| Attention heads | 32, 2 key/value heads |
| Context length | 262,144 tokens |
| Vocabulary | 131,072 tokens |
| Weight format | bfloat16 |
| Licence | bsd-2-clause |
| Base model | nvidia/NVIDIA-Nemotron-3-Nano-30B-A3B-BF16 |
Memory
Parameter count multiplied by the bytes each format uses. The key-value cache comes on top.
| Precision | Weights | Notes |
|---|---|---|
| bf16 | 58.8 GiB | Full precision as released |
| fp8 | 29.4 GiB | 8-bit, near-lossless on most models |
| int4 | 14.7 GiB | 4-bit, smallest footprint |
The cache costs 52 KiB per token at 16-bit, so the full 262,144-token context of one request needs about 13.0 GiB. Concurrency multiplies that number, not the weights.
Hardware
At the released precision, with 90% of the card given to the server and 2.5 GiB kept for the runtime. The context column is what is left for the cache on that setup.
| GPU | Cards needed | Free for cache | Context it holds |
|---|---|---|---|
| RTX 4090 24GB | 4 | 17.6 GiB | the full 262,144 tokens |
| RTX 5090 32GB | 4 | 46.4 GiB | the full 262,144 tokens |
| L40S 48GB | 2 | 22.6 GiB | the full 262,144 tokens |
| A100 80GB | 1 | 10.7 GiB | about 215,391 tokens |
| H100 80GB | 1 | 10.7 GiB | about 215,391 tokens |
| RTX PRO 6000 96GB | 1 | 25.1 GiB | the full 262,144 tokens |
| H200 141GB | 1 | 65.6 GiB | the full 262,144 tokens |
Serving
A vLLM launch line for the shape above. Check the model card for a runtime the publisher recommends.
vllm serve pipecat-ai/phonellm-alpha-1 \ --max-model-len 262144 \ --gpu-memory-utilization 0.90
Questions
The weights take about 59 GiB at the released precision. That fits on one A100 80GB, which leaves roughly 11 GiB for the key-value cache.
The repository declares bsd-2-clause. Read the licence text before commercial use: the name alone does not tell you what is allowed.
The configuration allows 262,144 tokens. Whether the whole window is usable depends on the memory left for the cache, which the hardware table works out per card.
This page is a reference, and Aiqre does not serve this model today. We deploy open-weight models on dedicated EU hardware on request, so if you want an OpenAI-compatible endpoint for it, email [email protected] with the model name and roughly what volume you expect.
Also in the library
Pantheon-Reasoning-26B-A4B-1.1-V2 is a 26.5 billion-parameter Gemma-4 mixture-of-experts finetune aimed at role-play scenarios, i…
Nex-N2.5-mini is a 35.1 billion-parameter mixture-of-experts language model released under Apache-2.0.
LLaDA2.2-mini is a 16.3 billion-parameter MoE diffusion language model with 128 k token context, supporting Levenshtein editing.
K2-Horizon-32B-Stage1 is a 32-billion-parameter dense decoder-only model released by IFM for benchmarking agentic, coding and rea…
Hosting
Tell us the model and roughly what volume you expect. We reply with a price and an OpenAI-compatible endpoint, with your prompts never stored.