SAVRN
Search Contact SAVRN

Open-weight model · Text generation

Ornith-1.5-35B-A3B-NVFP4

by Ornith ornith-ai/Ornith-1.5-35B-A3B-NVFP4

Ornith-1.5-35B-A3B-NVFP4 is an open-weight model for text generation from Ornith, released under MIT License. It has 19.5B parameters and a 262,144-token context. At 16-bit it needs about 46.9 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index. It draws 1.1M downloads a month.

Chirp Chirp! We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.

Parameters19.5B
Context262,144
Weights23.4 GB
Licensemit
AccessOpen weights
Monthly Downloads1.1M

Runs On

What it takes to serve Ornith-1.5-35B-A3B-NVFP4 (19.5B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.

PrecisionWeightsMemory neededCheapest setupPer hourAlso fits
16-bit 39.1 GB 46.9 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
8-bit 19.5 GB 23.4 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00
4-bit 9.8 GB 11.7 GB 1x MI300X (192 GB)
Vultr		 $1.85 1x H100 $1.99 · 1x MI325X $2.00

Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 10, 2026.

Ornith-1.5-35B-A3B-NVFP4 on every accelerator the SAVRN Index prices, at every precision

SAVRN's Notes on Ornith-1.5-35B-A3B-NVFP4

The name says 35B-A3B, the parameter field says 19.5B, and the published weights are 23.4 GB. All three can be true of a 256-expert mixture with 8 experts active per token, but it means confirming the precision you actually load before sizing hardware. Our table reads 46.9 GB needed at 16-bit, 23.4 GB at 8-bit and 11.7 GB at 4-bit — all fit one MI300X at $1.85 an hour, with the 262,144-token window the real consumer.

Ornith's pitch is process, not benchmarks: a self-improvement loop that generates its own tasks, builds scaffolds and rolls out solutions, refined with reinforcement learning, extending Ornith-1.0 — which the card says was built on Qwen3.5 and Gemma4 with continued pre-training and post-training. MIT makes commercial use open. At 1M downloads a month and with no base-model terms on the page, the homework is confirming upstream obligations yourself.

SAVRN Research

Model Card

By Ornith, published under mit, revision 94e431d9cc47.

Ornith-1.5-35B-A3B

Chirp Chirp! We are introducing Ornith-1.5, a major step toward building foundation models through end-to-end self-improvement.

Ornith-1.5 extends Ornith-1.0 (which was developed on top of Qwen3.5 and Gemma4 with additional continued pretraining, mid-training, and post-training) by expanding the self-improvement loop from scaffold and rollout optimization to jointly optimizing task generation, scaffold construction, and solution rollouts. Rather than relying on a fixed set of human-curated tasks and manually designed harnesses, Ornith-1.5 continuously generates new training tasks, discovers effective strategies for solving them, and improves the policy through reinforcement learning. For more details on the task, harness, and rollout reward design, please refer to our blog.

Ornith 1.5 35B-A3B

Read the full model card (1,476 words)

Configuration

Architecture
Qwen3_5MoeForConditionalGeneration
Context length (tokens)
262,144
Layers
40
Hidden size
2,048
Attention heads
16
Key/value heads
2
Head dimension
256
Vocabulary size
248,320
Experts
256
Experts active per token
8
Model type
qwen3_5_moe
Quantization
modelopt

Identity and Version

Repository
ornith-ai/Ornith-1.5-35B-A3B-NVFP4
Publisher
Ornith
Task
Text generation
Modality
Text
Library
transformers
Parameters
19.5B parameters
Languages
Not stated by the source
Revision
94e431d9cc47fa1986a7a1a4e9a80f7f118b03aa
First published
2026-08-18
Last updated
2026-08-26

Files and Weights

22 files, 23.5 GB in total. The weights are 3 files totalling 23.4 GB in safetensors.

Weights3 files · 23.4 GB
Configuration7 files · 10.5 MB
Tokenizer4 files · 22.9 MB
Documentation1 file · 29.9 KB
Other3 files · 1.7 MB
Repository4 files · 4.7 MB
Every file
FileTypeSizeSHA-256
model-00001-of-00003.safetensorsWeights10.0 GB 648fe8317148
model-00002-of-00003.safetensorsWeights10.0 GB cb5c5978721a
model-00003-of-00003.safetensorsWeights3.4 GB 8303cef7d375
config.jsonConfiguration58.2 KB —
generation_config.jsonConfiguration202 B —
hf_quant_config.jsonConfiguration42.2 KB —
model.safetensors.index.jsonConfiguration10.4 MB —
preprocessor_config.jsonConfiguration390 B —
processor_config.jsonConfiguration1.2 KB —
video_preprocessor_config.jsonConfiguration385 B —
README.mdDocumentation29.9 KB —
assets/ornith_35b_eval.pngOther709.4 KB 758a8f72b41a
assets/ornith_logo.pngOther962.4 KB 458ee0d85bae
chat_template.jinjaOther7.5 KB —
.gitattributesRepository1.8 KB —
.model_only_prunedRepository —
.quant_summary.txtRepository4.7 MB —
.sync_completeRepository295 B —
merges.txtTokenizer3.4 MB —
tokenizer.jsonTokenizer12.8 MB 5f9e4d4901a9
tokenizer_config.jsonTokenizer16.7 KB —
vocab.jsonTokenizer6.7 MB —

License and Download

License
mit
Access
Open weights, no gate
Download size
23.4 GB
Download from Ornith

Released by Ornith through its official repository on Hugging Face. Read the license.

Memory Requirements

PrecisionWeights in memory
As published23.4 GB
16-bit39.1 GB
8-bit19.5 GB
4-bit9.8 GB

Weights only, from the published parameter count; the key-value cache and runtime add to this.

Compare Ornith-1.5-35B-A3B-NVFP4

Questions About Ornith-1.5-35B-A3B-NVFP4

How much GPU memory does Ornith-1.5-35B-A3B-NVFP4 need?

About 46.9 GB at 16-bit and 11.7 GB at 4-bit: the weights (19.5B parameters) plus a working margin. A long context needs more.

What is the cheapest GPU to run Ornith-1.5-35B-A3B-NVFP4 on?

At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.

Can I use Ornith-1.5-35B-A3B-NVFP4 commercially?

Yes. Ornith-1.5-35B-A3B-NVFP4 is released under MIT License. The MIT License is a short permissive license. It permits commercial use, modification and redistribution, provided the copyright notice and permission notice are included.

What is Ornith-1.5-35B-A3B-NVFP4's context length?

262,144 tokens, from the maximum position embeddings in its published configuration.

Similar Models

An experimental NanoSeedLM (SeedLM) compression of IFM/K2-Horizon-MoVA-36B-A4B for Apple Silicon Macs with 32 GB of unified memory. SeedLM keeps each block of 8 weights as a 16-bit seed of a linear feedback shift register (LFSR), 4 coefficients and an exponent. The GPU makes the weights again from the seed at run time. The paper used an FPGA for this. This model uses Metal kernels on the Mac GPU. This is a research proof of concept. It is not a replacement for a calibrated quantization. - Six safetensors shards (22.76 GB in total, at most 4.5 GB each) and model.safetensors.index.json. The other files are config.json, the tokenizer files of the base model and nanoseedlmk2.py, the MLX loader.…

Open weights apache-2.0 19.2B parameters 524,288 tokens mlx
View model

Model · Text generation

Qwen3-32B-NVFP4-W4A4

Cezar

NVFP4 W4A4 quantization of (revision 9216db5781bf21249d130ec9da846c4624c16137, BF16). It was made for a controlled NVFP4-vs-MXFP4 decode benchmark on NVIDIA B200 (fp4bench), not as a general-purpose release: the NVFP4 and MXFP4 checkpoints share the model, the tool, the recipe, the calibration data and the layer coverage, and differ only in the format. Recipe: QuantizationModifier(targets="Linear", scheme="NVFP4", ignore=["lmhead"]). The plain preset recipe, with no weight-rounding optimization (GPTQ, AutoRound and the like). anon8231489123/ShareGPTVicunaunfiltered at revision 192ab2185289094fc556ec8ce5ce1e8e587154ca, at most 1024 tokens each (33208 tokens in all, seed 3). The data sets…

Open weights apache-2.0 19.1B parameters 40,960 tokens
View model

Model · Text generation

Qwen3.6-35B-A3B-NVFP4

NVIDIA

The NVIDIA Qwen3.6-35B-A3B-NVFP4 model is the quantized version of Alibaba's Qwen3.6-35B-A3B model, which is an auto-regressive language model that uses an optimized transformer architecture. For more information, please check here. The NVIDIA Qwen3.6-35B-A3B-NVFP4 model is quantized with Model Optimizer. This model is ready for commercial/non-commercial use. This model is not owned or developed by NVIDIA. This model has been developed and built to a third-party’s requirements for this application and use case; see link to Non-NVIDIA (Qwen3.6-35B-A3B) Model Card from Alibaba. Global Developers looking to take off-the-shelf, pre-quantized models for deployment in AI Agent systems, chatbots…

Open weights apache-2.0 18.7B parameters 262,144 tokens Model Optimizer
View model

Apache-2.0 This repository contains an NVFP4 quantization of VeriLoop E2, an open 27B post-trained model built on Qwen3.8-27B for code, mathematics, and physics. Its core reasoning discipline is VeriLoop-Governed Recurrence (VGR): candidate states are recursively proposed, externally checked, and retained only when the protected evidence state improves without regression. Quantized by Rodrigo Ramos. Produced with NVIDIA Model Optimizer using the canonical NVFP4W4A4WEIGHTLOCALHESSIANCFG recipe: static per-block 4-bit weights (group size 16) + dynamic 4-bit activations, FP8 attention, local-Hessian calibration with MSE and fp8 scale sweep. The linearattn (Gated Delta Net) blocks and the…

Open weights apache-2.0 18.3B parameters 262,144 tokens transformers
View model

Model · Text generation

gpt-neox-20b

EleutherAI

GPT-NeoX-20B is a 20 billion parameter autoregressive language model trained on the Pile using the GPT-NeoX library. Its architecture intentionally resembles that of GPT-3, and is almost identical to that of GPT-J- 6B. Its training dataset contains a multitude of English-language texts, reflecting the general-purpose nature of this model. See the accompanying paper for details about model architecture (including how it differs from GPT-3), training procedure, and additional evaluations. Model](https://arxiv.org/abs/2204.06745). For details about the training dataset, see the Pile paper, and its data sheet. Discord](https://discord.gg/zBGx3azzUn), and post them in #release-discussion. Please…

Open weights apache-2.0 20.7B parameters 2,048 tokens transformers
View model

Model · Text generation

NVIDIA-Nemotron-3-Nano-30B-A3B-NVFP4

NVIDIA

September 2025 \- December 2025 The post-training data has a cutoff date of November 28, 2025\. The pre-training data has a cutoff date of June 25, 2025\. Nemotron-Nano-3-30B-A3B-NVFP4 is a quantized version of Nemotron-Nano-3-30B-A3B and is a large language model (LLM) trained from scratch by NVIDIA, and designed as a unified model for both reasoning and non-reasoning tasks. It responds to user queries and tasks by first generating a reasoning trace and then concluding with a final response. The model's reasoning capabilities can be configured through a flag in the chat template. If the user prefers the model to provide its final answer without intermediate reasoning traces, it can be…

Open weights other 18.2B parameters 262,144 tokens transformers
View model