Qwen3.8-Flash-Next-MoEspressoV3 by Riccardo: Open Model

Model Card

This package brings Qwen3.8-Flash-Next to MoEspresso. Qwen3.8-Flash-Next is a 125B Mixture-of-Experts model with 6B active parameters per token and a 51B-parameter PLE/n-gram memory. On a 2021 M1 Max with a 24-core GPU and 32 GB of unified memory, MoEspresso serves this package at roughly 12–15 decode tokens/s. The local bounded configuration also scored 84.3% on a frozen 48-question generated-answer comparison, ahead of Claude Opus 4.8 xhigh at 80.3% on the same questions. The complete download occupies 145.75 GB (135.74 GiB), including the original BF16 PLE tables. MoEspresso keeps only its active working set in unified memory. The package requires MoEspresso 3.0.0 on an arm64 Apple…

Excerpt from the card by Riccardo, licensed other.

Configuration

Architecture: Qwen3_5_MoeForCausalLM
Context length (tokens): 262,144
Layers: 48
Hidden size: 2,560
Attention heads: 24
Key/value heads: 2
Head dimension: 256
Vocabulary size: 248,320
Experts: 512
Experts active per token: 10
Model type: qwen4_exp

Identity and Version

Repository: steadfastgaze/Qwen3.8-Flash-Next-MoEspressoV3
Publisher: Riccardo
Task: Text generation
Modality: Text
Library: mlx
Parameters: Not stated by the source
Languages: mlx
Revision: 7c830239d6ed14033b62801a7cbf13f3ef9da6f1
First published: 2026-09-15
Last updated: 2026-09-24

Files and Weights

196 files, 145.8 GB in total. The weights are 50 files totalling 43.4 GB in safetensors.

Weights50 files · 43.4 GB

Configuration10 files · 2.5 MB

Tokenizer4 files · 22.9 MB

Documentation2 files · 10.5 KB

Other129 files · 102.4 GB

Repository1 file · 9.4 KB

Every file

File	Type	Size	SHA-256
model-00001-of-00050.safetensors	Weights	4.3 GB	4ed743399514
model-00002-of-00050.safetensors	Weights	1.2 GB	fac45303ec54
model-00003-of-00050.safetensors	Weights	1.2 GB	032fdf059ad5
model-00004-of-00050.safetensors	Weights	796.9 MB	6ff7ef943acc
model-00005-of-00050.safetensors	Weights	796.9 MB	d49ac0f68ecb
model-00006-of-00050.safetensors	Weights	796.9 MB	d0328c64f335
model-00007-of-00050.safetensors	Weights	796.9 MB	86054a80083c
model-00008-of-00050.safetensors	Weights	796.9 MB	827fc9420c85
model-00009-of-00050.safetensors	Weights	796.9 MB	5ff393d10b36
model-00010-of-00050.safetensors	Weights	796.9 MB	14df1e1a3d53
model-00011-of-00050.safetensors	Weights	796.9 MB	a7e6f4a97044
model-00012-of-00050.safetensors	Weights	796.9 MB	45e339b4bb02
model-00013-of-00050.safetensors	Weights	796.9 MB	a572cc07ca84
model-00014-of-00050.safetensors	Weights	796.9 MB	59b9c564a0e1
model-00015-of-00050.safetensors	Weights	796.9 MB	59842fb75d3a
model-00016-of-00050.safetensors	Weights	796.9 MB	d680bbdc6a73
model-00017-of-00050.safetensors	Weights	796.9 MB	a32845bf22cb
model-00018-of-00050.safetensors	Weights	796.9 MB	3717c80d2d49
model-00019-of-00050.safetensors	Weights	796.9 MB	b2ca0647fd52
model-00020-of-00050.safetensors	Weights	796.9 MB	28b42f464a7e
model-00021-of-00050.safetensors	Weights	796.9 MB	6935d1b9bcac
model-00022-of-00050.safetensors	Weights	796.9 MB	ad33604201bf
model-00023-of-00050.safetensors	Weights	796.9 MB	5e4d611734bc
model-00024-of-00050.safetensors	Weights	796.9 MB	45f134ad0cab
model-00025-of-00050.safetensors	Weights	796.9 MB	08a358a088a4
model-00026-of-00050.safetensors	Weights	796.9 MB	8a03f343becc
model-00027-of-00050.safetensors	Weights	796.9 MB	2de2371e7dd0
model-00028-of-00050.safetensors	Weights	796.9 MB	81a4536487f0
model-00029-of-00050.safetensors	Weights	796.9 MB	dcae5e77b6cb
model-00030-of-00050.safetensors	Weights	796.9 MB	564f5141b256
model-00031-of-00050.safetensors	Weights	796.9 MB	001988478d74
model-00032-of-00050.safetensors	Weights	796.9 MB	1251c2567d2b
model-00033-of-00050.safetensors	Weights	796.9 MB	d4c5ae2fe537
model-00034-of-00050.safetensors	Weights	796.9 MB	adbf6ca4a5bc
model-00035-of-00050.safetensors	Weights	796.9 MB	bf4b17105422
model-00036-of-00050.safetensors	Weights	796.9 MB	537a5737f3d2
model-00037-of-00050.safetensors	Weights	796.9 MB	75308335f310
model-00038-of-00050.safetensors	Weights	796.9 MB	aa2acf1c74fb
model-00039-of-00050.safetensors	Weights	796.9 MB	1c4c826fa516
model-00040-of-00050.safetensors	Weights	796.9 MB	a7928d357a0d
model-00041-of-00050.safetensors	Weights	796.9 MB	543c67550a0f
model-00042-of-00050.safetensors	Weights	796.9 MB	8628e8f7412a
model-00043-of-00050.safetensors	Weights	796.9 MB	28b47cf6c638
model-00044-of-00050.safetensors	Weights	796.9 MB	cf0b75f04f4a
model-00045-of-00050.safetensors	Weights	796.9 MB	b79be00b50ec
model-00046-of-00050.safetensors	Weights	796.9 MB	c52296484cea
model-00047-of-00050.safetensors	Weights	796.9 MB	6cbfc104199f
model-00048-of-00050.safetensors	Weights	796.9 MB	47e2868212d1
model-00049-of-00050.safetensors	Weights	796.9 MB	aae0d6cfa34e
model-00050-of-00050.safetensors	Weights	131.0 MB	0c3ca3df5091
config.json	Configuration	3.6 KB	—
expert_hotlist.json	Configuration	320.3 KB	—
generation_config.json	Configuration	202 B	—
jang_config.json	Configuration	488 B	—
package_manifest.json	Configuration	901.5 KB	—
package_plan.json	Configuration	657.1 KB	—
preprocessor_config.json	Configuration	390 B	—
qwen4_iqk_package_report.json	Configuration	7.9 KB	—
source_inventory.json	Configuration	656.3 KB	—
video_preprocessor_config.json	Configuration	385 B	—
LICENSE	Documentation	3.2 KB	—
README.md	Documentation	7.3 KB	—
chat_template.jinja	Other	9.0 KB	—
ple/rows-000-of-128.bf16	Other	800.0 MB	f94d3ac0a76d
ple/rows-001-of-128.bf16	Other	800.0 MB	2382696fc117
ple/rows-002-of-128.bf16	Other	800.0 MB	28b4ac76b117
ple/rows-003-of-128.bf16	Other	800.0 MB	31fdedfe70c6
ple/rows-004-of-128.bf16	Other	800.0 MB	ab6a7f5f878b
ple/rows-005-of-128.bf16	Other	800.0 MB	3a9cd48fb062
ple/rows-006-of-128.bf16	Other	800.0 MB	d747fc34b152
ple/rows-007-of-128.bf16	Other	800.0 MB	06bead6e0708
ple/rows-008-of-128.bf16	Other	800.0 MB	f79b9f6013c4
ple/rows-009-of-128.bf16	Other	800.0 MB	5bde28146b7c
ple/rows-010-of-128.bf16	Other	800.0 MB	db76c5dad4a7
ple/rows-011-of-128.bf16	Other	800.0 MB	97110fddbd26
ple/rows-012-of-128.bf16	Other	800.0 MB	18bc101ba3e9
ple/rows-013-of-128.bf16	Other	800.0 MB	700995416da1
ple/rows-014-of-128.bf16	Other	800.0 MB	fc8b2cae83be
ple/rows-015-of-128.bf16	Other	800.0 MB	f32e09d1e06a
ple/rows-016-of-128.bf16	Other	800.0 MB	2b198105476a
ple/rows-017-of-128.bf16	Other	800.0 MB	a52182c95e64
ple/rows-018-of-128.bf16	Other	800.0 MB	79326f35e698
ple/rows-019-of-128.bf16	Other	800.0 MB	a5296d8b9962
ple/rows-020-of-128.bf16	Other	800.0 MB	3c69fe8995d4
ple/rows-021-of-128.bf16	Other	800.0 MB	35b00bea4b8b
ple/rows-022-of-128.bf16	Other	800.0 MB	c7a1c9127e2c
ple/rows-023-of-128.bf16	Other	800.0 MB	9251c2ab5d59
ple/rows-024-of-128.bf16	Other	800.0 MB	534b830f2bc0
ple/rows-025-of-128.bf16	Other	800.0 MB	370076308815
ple/rows-026-of-128.bf16	Other	800.0 MB	b04c206609d5
ple/rows-027-of-128.bf16	Other	800.0 MB	1e96f12d5fe4
ple/rows-028-of-128.bf16	Other	800.0 MB	7f32d02eaf96
ple/rows-029-of-128.bf16	Other	800.0 MB	57d1b47729f5
ple/rows-030-of-128.bf16	Other	800.0 MB	46d6a59903de
ple/rows-031-of-128.bf16	Other	800.0 MB	b8be75e23c05
ple/rows-032-of-128.bf16	Other	800.0 MB	96a2a20ef907
ple/rows-033-of-128.bf16	Other	800.0 MB	5e76cafd4358
ple/rows-034-of-128.bf16	Other	800.0 MB	7d1b5a916091
ple/rows-035-of-128.bf16	Other	800.0 MB	4faec9010aac
ple/rows-036-of-128.bf16	Other	800.0 MB	dadd29c0c3b3
ple/rows-037-of-128.bf16	Other	800.0 MB	227e912e6cbe
ple/rows-038-of-128.bf16	Other	800.0 MB	58de8d878e3b
ple/rows-039-of-128.bf16	Other	800.0 MB	17cd7a0bd5bc
ple/rows-040-of-128.bf16	Other	800.0 MB	05ea8fb526d1
ple/rows-041-of-128.bf16	Other	800.0 MB	b3b0b33f03a8
ple/rows-042-of-128.bf16	Other	800.0 MB	d601c2d1eda0
ple/rows-043-of-128.bf16	Other	800.0 MB	4df3748ce7aa
ple/rows-044-of-128.bf16	Other	800.0 MB	e7d6479fa5e4
ple/rows-045-of-128.bf16	Other	800.0 MB	445225bb371c
ple/rows-046-of-128.bf16	Other	800.0 MB	0ec704e0056d
ple/rows-047-of-128.bf16	Other	800.0 MB	31570f1520ae
ple/rows-048-of-128.bf16	Other	800.0 MB	7ae1dada68c0
ple/rows-049-of-128.bf16	Other	800.0 MB	b6599691fbcf
ple/rows-050-of-128.bf16	Other	800.0 MB	ead2ff2ab0f5
ple/rows-051-of-128.bf16	Other	800.0 MB	866de2b62fb1
ple/rows-052-of-128.bf16	Other	800.0 MB	bdca123754d0
ple/rows-053-of-128.bf16	Other	800.0 MB	67c78ef24a2d
ple/rows-054-of-128.bf16	Other	800.0 MB	87af18ee7850
ple/rows-055-of-128.bf16	Other	800.0 MB	f0346286597d
ple/rows-056-of-128.bf16	Other	800.0 MB	061a486edb66
ple/rows-057-of-128.bf16	Other	800.0 MB	cc9af1e3f310
ple/rows-058-of-128.bf16	Other	800.0 MB	29ad2c0b7d10
ple/rows-059-of-128.bf16	Other	800.0 MB	cad5bf6a36b7
ple/rows-060-of-128.bf16	Other	800.0 MB	0a7871c6e02a
ple/rows-061-of-128.bf16	Other	800.0 MB	ab60cb6ca258
ple/rows-062-of-128.bf16	Other	800.0 MB	9b1559ec616d
ple/rows-063-of-128.bf16	Other	800.0 MB	3387b6f0fd52
ple/rows-064-of-128.bf16	Other	800.0 MB	21b372c328c8
ple/rows-065-of-128.bf16	Other	800.0 MB	c7a63f0c8804
ple/rows-066-of-128.bf16	Other	800.0 MB	033955ca0e48
ple/rows-067-of-128.bf16	Other	800.0 MB	dd587d560920
ple/rows-068-of-128.bf16	Other	800.0 MB	6f94d690d921
ple/rows-069-of-128.bf16	Other	800.0 MB	82716e358866
ple/rows-070-of-128.bf16	Other	800.0 MB	49bc6c8a1f20
ple/rows-071-of-128.bf16	Other	800.0 MB	b132e24f489a
ple/rows-072-of-128.bf16	Other	800.0 MB	1ee93f25742d
ple/rows-073-of-128.bf16	Other	800.0 MB	bab7e5b4e7ef
ple/rows-074-of-128.bf16	Other	800.0 MB	0a7835b7da93
ple/rows-075-of-128.bf16	Other	800.0 MB	d276ebc58970
ple/rows-076-of-128.bf16	Other	800.0 MB	4c573441abfa
ple/rows-077-of-128.bf16	Other	800.0 MB	505cbd094f9f
ple/rows-078-of-128.bf16	Other	800.0 MB	8701cd7f24eb
ple/rows-079-of-128.bf16	Other	800.0 MB	60113fd9ee0f
ple/rows-080-of-128.bf16	Other	800.0 MB	609e0e8d0e6d
ple/rows-081-of-128.bf16	Other	800.0 MB	361a5ff2886d
ple/rows-082-of-128.bf16	Other	800.0 MB	63d11710494f
ple/rows-083-of-128.bf16	Other	800.0 MB	ebc043cb5aa1
ple/rows-084-of-128.bf16	Other	800.0 MB	411b098613bc
ple/rows-085-of-128.bf16	Other	800.0 MB	d3a45675ff62
ple/rows-086-of-128.bf16	Other	800.0 MB	fe9a6c2a866c
ple/rows-087-of-128.bf16	Other	800.0 MB	278a02d1f7ee
ple/rows-088-of-128.bf16	Other	800.0 MB	13398731df18
ple/rows-089-of-128.bf16	Other	800.0 MB	3a1711ef13bf
ple/rows-090-of-128.bf16	Other	800.0 MB	b574e3783c30
ple/rows-091-of-128.bf16	Other	800.0 MB	814146ad1405
ple/rows-092-of-128.bf16	Other	800.0 MB	edfae581f12f
ple/rows-093-of-128.bf16	Other	800.0 MB	eb72e4a33049
ple/rows-094-of-128.bf16	Other	800.0 MB	e68fa5222c52
ple/rows-095-of-128.bf16	Other	800.0 MB	59986510acc1
ple/rows-096-of-128.bf16	Other	800.0 MB	5e3df7170bc4
ple/rows-097-of-128.bf16	Other	800.0 MB	d0c8151982e2
ple/rows-098-of-128.bf16	Other	800.0 MB	81f7d166a3f9
ple/rows-099-of-128.bf16	Other	800.0 MB	15018db1b5b1
ple/rows-100-of-128.bf16	Other	800.0 MB	97e4d075478f
ple/rows-101-of-128.bf16	Other	800.0 MB	d0e12fe85e1c
ple/rows-102-of-128.bf16	Other	800.0 MB	939214daa6ff
ple/rows-103-of-128.bf16	Other	800.0 MB	1185fd41832a
ple/rows-104-of-128.bf16	Other	800.0 MB	1c8fc2a7a6a9
ple/rows-105-of-128.bf16	Other	800.0 MB	1333cff8ef54
ple/rows-106-of-128.bf16	Other	800.0 MB	5c2a2a9a3cb8
ple/rows-107-of-128.bf16	Other	800.0 MB	df62e70bfaaf
ple/rows-108-of-128.bf16	Other	800.0 MB	c75c7f5cc2bb
ple/rows-109-of-128.bf16	Other	800.0 MB	304f74528633
ple/rows-110-of-128.bf16	Other	800.0 MB	5bc6f69a2484
ple/rows-111-of-128.bf16	Other	800.0 MB	7c9caff71fc1
ple/rows-112-of-128.bf16	Other	800.0 MB	1561a9fe0f15
ple/rows-113-of-128.bf16	Other	800.0 MB	acbe13a32b90
ple/rows-114-of-128.bf16	Other	800.0 MB	48d67bf6d7f9
ple/rows-115-of-128.bf16	Other	800.0 MB	2dc549699361
ple/rows-116-of-128.bf16	Other	800.0 MB	6961df56d83e
ple/rows-117-of-128.bf16	Other	800.0 MB	8ed0ae29b3a1
ple/rows-118-of-128.bf16	Other	800.0 MB	c1ed984649e8
ple/rows-119-of-128.bf16	Other	800.0 MB	3d5ed304889c
ple/rows-120-of-128.bf16	Other	800.0 MB	c9bec8843e21
ple/rows-121-of-128.bf16	Other	800.0 MB	c652729ac051
ple/rows-122-of-128.bf16	Other	800.0 MB	704015602434
ple/rows-123-of-128.bf16	Other	800.0 MB	2eb38b64efd8
ple/rows-124-of-128.bf16	Other	800.0 MB	48a34ab6a94b
ple/rows-125-of-128.bf16	Other	800.0 MB	0acc6a42a5c5
ple/rows-126-of-128.bf16	Other	800.0 MB	0e5be27b68a0
ple/rows-127-of-128.bf16	Other	800.0 MB	6252be6115ec
.gitattributes	Repository	9.4 KB	—
merges.txt	Tokenizer	3.4 MB	—
tokenizer.json	Tokenizer	12.8 MB	0997f410c57a
tokenizer_config.json	Tokenizer	17.9 KB	—
vocab.json	Tokenizer	6.7 MB	—

License and Download

License: other
Access: Open weights, no gate
Download size: 43.4 GB

Download from Riccardo

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

Built From

Derived from Qwen/Qwen3.8-Flash-Next
Quantized from Qwen/Qwen3.8-Flash-Next

Memory Requirements

Precision	Weights in memory
As published	43.4 GB

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

Questions About Qwen3.8-Flash-Next-MoEspressoV3

What license is Qwen3.8-Flash-Next-MoEspressoV3 released under?

other, as its publisher declares it. Read the license text before commercial use.

What is Qwen3.8-Flash-Next-MoEspressoV3's context length?

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

Similar Models

Model · Text generation

opt-125m

AI at Meta

OPT was first introduced in Open Pre-trained Transformer Language Models and first released in metaseq's repository on May 3rd 2022 by Meta AI. Disclaimer: The team releasing OPT wrote an official model card, which is available in Appendix D of the paper. Content from this model card has been written by the Hugging Face team. To quote the first two paragraphs of the official paper OPT was predominantly pretrained with English text, but a small amount of non-English data is still present within the training corpus via CommonCrawl. The model was pretrained using a causal language modeling (CLM) objective. OPT belongs to the same family of decoder-only models like GPT-3. As such, it was…

Open weights other 2,048 tokens transformers

View model

Model · Text generation

Ornith-1.5-9B-GGUF

Ornith

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…

Open weights mit transformers

View model

Model · Text generation

Ternary-Bonsai-2-27B-gguf

Prism ML

Full 27B-class reasoning in ternary transformer weights, for llama.cpp (CUDA, Metal, CPU) - \~5.9 GB language model (down from \~54 GB FP16) — full 27B-class reasoning on a standard laptop or a single GPU - 98.2% of FP16 intelligence retained: 84.78 average across 14 thinking-mode benchmarks — far above the conventional IQ2XXS build (72.59) at about 82% of its footprint, and within 0.4 points of UD-Q4KXL at three times the footprint - Retains thinking, reasoning, and agentic behavior deep in the sub-4-bit regime, where conventional low-bit representations collapse: math within half a point of full precision (96.57), coding level with the baseline (89.42), agentic tool calling at 74.92…

Open weights apache-2.0 llama.cpp

View model

Model · Text generation

Ornith-1.5-35B-A3B-GGUF

Ornith

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…

Open weights mit transformers

View model

Model · Text generation

Qwen3.8-27B-Uncensored-GGUF

Jonathan Coletti

Uncensored Qwen3.8-27B, published as GGUF quantizations with the multi token prediction (MTP) head retained and verified. Refusal behaviour has been substantially reduced, not eliminated. See Measured behaviour for the numbers. Capabilities, training data, and architecture are otherwise unchanged. - Refusal directions removed with Heretic, which co minimizes refusal count against KL divergence from the base model. No handwritten refusal removal code, no finetuning, no additional training data. - Abliteration runs at bf16 (no 4 bit quantization). the resulting LoRA is merged into the bf16 base, so the published weights are not a quantized round trip. - mtp. tensors are copied verbatim from…

Open weights apache-2.0 llama.cpp

View model

Model · Text generation

Ornith-1.0-9B-GGUF

Ornith

Aloha! Today, we are releasing Ornith-1.0, a self-improving family of open-source models for agentic coding. This model card documents Ornith-1.0-9B, the most lightweight member of the Ornith family, designed for efficient single-GPU deployment. Ornith-1.0-9B is a dense ~9B model (≈19 GB in bf16), so it serves comfortably on a single 80GB GPU. The recipes below stand up an OpenAI-compatible server; add --tensor-parallel-size / --tp if you want to shard across more GPUs. For a quick local test (or to script offline generation), load the model directly with Transformers. Make sure you have a recent release installed — see the Transformers installation guide; Ornith-1.0-9B requires…

Open weights mit transformers

View model