Get the app
← back to catalog · registered 2026-11-03 11:02

gatesushi/Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored-oQ4e-mtp

gatesushi 35B MoE
Your rig guess connected
? Why do I need an app?
Reading your rig…

This is a rough estimate. Install the free app - we'll show exact numbers.

Reading real hardware from your app right now. Numbers below are exact.

Below is the per-quantization compatibility for this model. 

curl -H "Authorization: Bearer $ABL_KEY" \
     "https://abliteration.org/api/v1/models/gatesushi%2FQwen3.6-35B-A3B-Brainwaves-Nex-Uncensored-oQ4e-mtp"
Response includes
  • classification m-uncensored
  • files 14
  • author_summary 1 models
  • readme_text full
Try next
GET/methods/m-uncensored
What "m-uncensored" means · taxonomy
10 credits · hourly refresh · ~4 KB payload Get an API key →
Abliteration classifier · v1.0.0
M-U
Primary method

Uncensored (method unknown)

No other method signals detected in this model.
Confidence
LOW
Why this label 3 signals
Weak or ambiguous signals. Best guess based on catalog patterns; treat as tentative and check the evidence below.
  • 'uncensored' in name/tags but no 'abliterated' marker
  • method not identifiable from author declaration alone
  • may be DPO fine-tune, prompt engineering, or unknown technique
How we label → About M-U →
Refusal direction extraction

No specific extraction method could be identified for this model. The producer either did not document it or used a proprietary pipeline.

What is a refusal direction? →
Downloads · 30-day
0
Likes
0
Model age
today
created 2026-11-03

Metadata

Tags
transformers safetensors qwen3_5_moe image-text-to-text unsloth qwen3_5 qwen qwen3.5 distillation reasoning chain-of-thought long-cot
Total size
20.6 GB
Files
14
Quantizations
1
Registered
2026-11-03 11:02
Last updated on HF
2026-11-03 11:44
Source huggingface.co/gatesushi/Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored-oQ4e-mtp ↗

Files by quantization

Auxiliary files 14 files 20.6 GB
model-00003-of-00005.safetensors 4.78 GB 133ab4ba download
model-00004-of-00005.safetensors 4.78 GB 9dcd8ef8 download
model-00002-of-00005.safetensors 4.78 GB cd4a3eb0 download
model-00001-of-00005.safetensors 4.72 GB 4a118440 download
model-00005-of-00005.safetensors 1.55 GB 351f2f35 download
tokenizer.json 19.1 MB 87a7830d download
model.safetensors.index.json 200 KB 8503b399 download
config.json 97.8 KB c68d726a download
oq_imatrix_report.json 32.0 KB 519cf028 download
README.md 15.3 KB 9138ea8a download
chat_template.jinja 7.56 KB 34398e43 download
.gitattributes 1.53 KB 52373fe2 download
tokenizer_config.json 1.11 KB a068e246 download
preprocessor_config.json 390 B 2ea84a43 download

README current version from Hugging Face

library_name: mlx
tags:

  • mlx
  • oq
  • quantized
  • mtp

Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored-oQ4e-mtp is an oQ4e quantized version of Nightmedia's model (Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored), a 4-bit-equivalent quantization of the Qwen3.6-35B-A3B architecture, deployed via the 0.7.3 oMLX framework using the splash engine.

Based on benchmarking against its higher-precision siblings (oQ6e-mtp and Nightmedia's own qx64-hi-mlx), this model emerges as an excellent daily-driver and general-purpose assistant, offering near-parity or superiority on most general-knowledge, math, and basic-coding tasks while generating responses significantly faster than the regular-system qx64-hi-mlx variant.

Performance & Speed

The oQ4e-mtp model decodes tokens roughly 2.9× faster than the qx64-hi-mlx model at standard conversation lengths and remains ~70% faster even at maximum 200k-token contexts.

Speed Comparison Snapshot

Metric oQ4e-mtp oQ6e-mtp qx64-hi-mlx
Peak Memory Usage ~22 – 25.6 GB ~29.8 – 33.4 GB ~29.3 – 33.6 GB
Initial Decode Speed (TG TPS @ pp1k) ~319 tok/s ~209 tok/s ~110 tok/s
Long-Context Decode Speed (TG TPS @ pp200k) ~84 tok/s ~78 tok/s ~49 tok/s
Max Context Throughput (Overall @ pp200k) ~1,191 tok/s ~1,173 tok/s ~1,278 tok/s

Stability & oMLX + Splash Variance Note

Benchmark runs for the splash-engine (1.5.0) models (oQ4e-mtp and oQ6e-mtp) show run-to-run timing variance that is spiky rather than uniform. It appears to be caused by something in the oMLX + splash engine combination that triggers at certain context window sizes (not seen in oMLX engine with qx64-hi-mlx).

  • Decode stability: Despite prefill timing swings, token generation speed (TG TPS) remains highly stable across all contexts, meaning day-to-day conversational responsiveness is unaffected.

For daily driving, these prefill/timing variances are benign measurement artifacts that do not impact answer quality or generation flow.

Intelligence & Benchmark Profile

On cognitive benchmarks, oQ4e-mtp matches or exceeds the qx64-hi-mlx model on the majority of categories, particularly in general knowledge, standardized testing, science QA, and math reasoning. Its primary trade-offs are nuanced common-sense resolution (Winogrande) and advanced live coding (LiveCodeBench), where the higher-tier models hold a modest edge.

Benchmark Comparison Snapshot

Benchmark oQ4e-mtp oQ6e-mtp qx64-hi-mlx
MMLU (General Knowledge) 87.0% 86.0% 86.0%
MMLU_PRO (Hard Reasoning) 67.0% 61.0% 62.0%
HELLASWAG (Common Sense) 95.5% 95.5% 93.0%
TRUTHFULQA 87.3% 88.0% 87.0%
ARC_CHALLENGE (Science QA) 97.3% 96.3% 95.7%
WINOGRANDE (Nuanced Pronouns) 76.3% 78.0% 81.3%
GSM8K (Math Word Problems) 96.0% 94.0% 95.0%
MATHQA (With Thinking Enabled) 90.5% 90.0% 89.5%
HUMANEVAL (Basic Coding) 94.0% 94.0% 94.0%
MBPP (Code Generation) 81.0% 78.0% 80.0%
LIVECODEBENCH (Advanced Live Coding) 58.0% 69.0% 67.0%

Recommended Use Cases & Daily Driver Verdict

oQ4e-mtp is an ideal daily-driver and general-model choice if you:

  • Prioritize fast, responsive text generation over absolute benchmark reproducibility.
  • Want strong all-around intelligence for writing, research, learning, and everyday coding tasks.

Example Capability Demonstration

With proper tuning of the settings and guidance from the system prompt, it is able to solve problems well.

This is the rendering of the SVG below

Screenshot 2026-11-03 at 4.42.54

Configuration Used

Temperature: 0.6
Max Tokens: Default
Top P: 0.95
Top K: 20
Min P: 0.05
Rep. Penalty: 1.1
Presence Penalty: 1.2


Given this PGN string of a chess game:

b3 e5 2. Nf3 h5 3. d4 exd4 4. Nxd4 Nf6 5. f4 Ke7 6. Qd3 d5 7. h4 *

Figure out the current state of the chessboard, and highlight the last move. Provide the XML text for an SVG image.
Show it from the point of view of whites.
After the SVG, also provide a plain-text board grid


Step 0: Feasibility Check

Goal: Parse the provided PGN, determine the current board state, highlight the last move, and produce an SVG board (from White’s perspective) plus a plain-text grid.
Constraints & Facts:

    PGN string: b3 e5 2. Nf3 h5 3. d4 exd4 4. Nxd4 Nf6 5. f4 Ke7 6. Qd3 d5 7. h4 *
    The opening lacks an explicit “1.”, but the sequence clearly implies: 1. b3 e5, followed by moves 2–7.
    The * at the end indicates the game is finished or unresolved; Black has no 7th move recorded.
    Last move: 7. h4 (White pawn h2 → h4).
    Perspective: White’s view (Rank 8 at top, Rank 1 at bottom, a-file left, h-file right).

Verdict: FEASIBLE. All moves are legal and lead to a unique, verifiable board state with no contradictions.
Parsed Move Sequence & Board State

    b3 (White pawn b2→b3) | e5 (Black pawn e7→e5)
    Nf3 (White Ng1→f3) | h5 (Black pawn h7→h5)
    d4 (White pawn d2→d4) | exd4 (Black pawn e5 captures d4)
    Nxd4 (White Knight f3 captures d4) | Nf6 (Black Ng8→f6)
    f4 (White pawn f2→f4) | Ke7 (Black King e8→e7)
    Qd3 (White Queen d1→d3) | d5 (Black pawn d7→d5)
    h4 (White pawn h2→h4) — Last move

Material lost: White lost the d2-pawn; Black lost the e7-pawn. All other pieces remain on their starting files except those explicitly moved.
SVG Image (XML)

<svg xmlns="http://www.w3.org/2000/svg" width="720" height="720" viewBox="0 0 720 720">
  <defs>
    <style>
      .light { fill: #f0d9b5; }
      .dark { fill: #b58863; }
      .highlight { fill: rgba(255, 255, 0, 0.55); }
      .piece { font-family: "Segoe UI Symbol", "Arial Unicode MS", sans-serif; font-size: 50px; text-anchor: middle; dominant-baseline: central; }
      .white-piece { fill: #ffffff; stroke: #000000; stroke-width: 1.2px; }
      .black-piece { fill: #000000; }
      .label { font-family: Arial, sans-serif; font-size: 16px; text-anchor: middle; dominant-baseline: central; fill: #dddddd; }
    </style>
  </defs>

  <!-- Outer background -->
  <rect x="0" y="0" width="720" height="720" fill="#333333"/>

  <!-- Board group -->
  <g transform="translate(40, 40)">
    <!-- Squares: Rank 8 -->
    <rect x="0" y="0" width="80" height="80" class="light"/>
    <rect x="80" y="0" width="80" height="80" class="dark"/>
    <rect x="160" y="0" width="80" height="80" class="light"/>
    <rect x="240" y="0" width="80" height="80" class="dark"/>
    <rect x="320" y="0" width="80" height="80" class="light"/>
    <rect x="400" y="0" width="80" height="80" class="dark"/>
    <rect x="480" y="0" width="80" height="80" class="light"/>
    <rect x="560" y="0" width="80" height="80" class="dark"/>

    <!-- Squares: Rank 7 -->
    <rect x="0" y="80" width="80" height="80" class="dark"/>
    <rect x="80" y="80" width="80" height="80" class="light"/>
    <rect x="160" y="80" width="80" height="80" class="dark"/>
    <rect x="240" y="80" width="80" height="80" class="light"/>
    <rect x="320" y="80" width="80" height="80" class="dark"/>
    <rect x="400" y="80" width="80" height="80" class="light"/>
    <rect x="480" y="80" width="80" height="80" class="dark"/>
    <rect x="560" y="80" width="80" height="80" class="light"/>

    <!-- Squares: Rank 6 -->
    <rect x="0" y="160" width="80" height="80" class="light"/>
    <rect x="80" y="160" width="80" height="80" class="dark"/>
    <rect x="160" y="160" width="80" height="80" class="light"/>
    <rect x="240" y="160" width="80" height="80" class="dark"/>
    <rect x="320" y="160" width="80" height="80" class="light"/>
    <rect x="400" y="160" width="80" height="80" class="dark"/>
    <rect x="480" y="160" width="80" height="80" class="light"/>
    <rect x="560" y="160" width="80" height="80" class="dark"/>

    <!-- Squares: Rank 5 -->
    <rect x="0" y="240" width="80" height="80" class="dark"/>
    <rect x="80" y="240" width="80" height="80" class="light"/>
    <rect x="160" y="240" width="80" height="80" class="dark"/>
    <rect x="240" y="240" width="80" height="80" class="light"/>
    <rect x="320" y="240" width="80" height="80" class="dark"/>
    <rect x="400" y="240" width="80" height="80" class="light"/>
    <rect x="480" y="240" width="80" height="80" class="dark"/>
    <rect x="560" y="240" width="80" height="80" class="light"/>

    <!-- Squares: Rank 4 -->
    <rect x="0" y="320" width="80" height="80" class="light"/>
    <rect x="80" y="320" width="80" height="80" class="dark"/>
    <rect x="160" y="320" width="80" height="80" class="light"/>
    <rect x="240" y="320" width="80" height="80" class="dark"/>
    <rect x="320" y="320" width="80" height="80" class="light"/>
    <rect x="400" y="320" width="80" height="80" class="dark"/>
    <rect x="480" y="320" width="80" height="80" class="light"/>
    <rect x="560" y="320" width="80" height="80" class="dark"/>

    <!-- Squares: Rank 3 -->
    <rect x="0" y="400" width="80" height="80" class="dark"/>
    <rect x="80" y="400" width="80" height="80" class="light"/>
    <rect x="160" y="400" width="80" height="80" class="dark"/>
    <rect x="240" y="400" width="80" height="80" class="light"/>
    <rect x="320" y="400" width="80" height="80" class="dark"/>
    <rect x="400" y="400" width="80" height="80" class="light"/>
    <rect x="480" y="400" width="80" height="80" class="dark"/>
    <rect x="560" y="400" width="80" height="80" class="light"/>

    <!-- Squares: Rank 2 -->
    <rect x="0" y="480" width="80" height="80" class="light"/>
    <rect x="80" y="480" width="80" height="80" class="dark"/>
    <rect x="160" y="480" width="80" height="80" class="light"/>
    <rect x="240" y="480" width="80" height="80" class="dark"/>
    <rect x="320" y="480" width="80" height="80" class="light"/>
    <rect x="400" y="480" width="80" height="80" class="dark"/>
    <rect x="480" y="480" width="80" height="80" class="light"/>
    <rect x="560" y="480" width="80" height="80" class="dark"/>

    <!-- Squares: Rank 1 -->
    <rect x="0" y="560" width="80" height="80" class="dark"/>
    <rect x="80" y="560" width="80" height="80" class="light"/>
    <rect x="160" y="560" width="80" height="80" class="dark"/>
    <rect x="240" y="560" width="80" height="80" class="light"/>
    <rect x="320" y="560" width="80" height="80" class="dark"/>
    <rect x="400" y="560" width="80" height="80" class="light"/>
    <rect x="480" y="560" width="80" height="80" class="dark"/>
    <rect x="560" y="560" width="80" height="80" class="light"/>

    <!-- Highlight last move: h2 (source) and h4 (destination) -->
    <rect x="560" y="480" width="80" height="80" class="highlight"/>
    <rect x="560" y="320" width="80" height="80" class="highlight"/>

    <!-- White Pieces -->
    <text class="piece white-piece" x="40" y="520">♙</text> <!-- a2 -->
    <text class="piece white-piece" x="120" y="440">♙</text> <!-- b3 -->
    <text class="piece white-piece" x="200" y="520">♙</text> <!-- c2 -->
    <text class="piece white-piece" x="280" y="360">♘</text> <!-- d4 -->
    <text class="piece white-piece" x="360" y="520">♙</text> <!-- e2 -->
    <text class="piece white-piece" x="440" y="360">♙</text> <!-- f4 -->
    <text class="piece white-piece" x="520" y="520">♙</text> <!-- g2 -->
    <text class="piece white-piece" x="600" y="360">♙</text> <!-- h4 -->
    <text class="piece white-piece" x="40" y="600">♖</text> <!-- a1 -->
    <text class="piece white-piece" x="120" y="600">♘</text> <!-- b1 -->
    <text class="piece white-piece" x="200" y="600">♗</text> <!-- c1 -->
    <text class="piece white-piece" x="280" y="440">♕</text> <!-- d3 -->
    <text class="piece white-piece" x="360" y="600">♔</text> <!-- e1 -->
    <text class="piece white-piece" x="440" y="600">♗</text> <!-- f1 -->
    <text class="piece white-piece" x="600" y="600">♖</text> <!-- h1 -->

    <!-- Black Pieces -->
    <text class="piece black-piece" x="40" y="40">♜</text> <!-- a8 -->
    <text class="piece black-piece" x="120" y="40">♞</text> <!-- b8 -->
    <text class="piece black-piece" x="200" y="40">♝</text> <!-- c8 -->
    <text class="piece black-piece" x="280" y="40">♛</text> <!-- d8 -->
    <text class="piece black-piece" x="360" y="120">♚</text> <!-- e7 -->
    <text class="piece black-piece" x="440" y="40">♝</text> <!-- f8 -->
    <text class="piece black-piece" x="600" y="40">♜</text> <!-- h8 -->
    <text class="piece black-piece" x="40" y="120">♟</text> <!-- a7 -->
    <text class="piece black-piece" x="120" y="120">♟</text> <!-- b7 -->
    <text class="piece black-piece" x="200" y="120">♟</text> <!-- c7 -->
    <text class="piece black-piece" x="440" y="120">♟</text> <!-- f7 -->
    <text class="piece black-piece" x="520" y="120">♟</text> <!-- g7 -->
    <text class="piece black-piece" x="440" y="200">♞</text> <!-- f6 -->
    <text class="piece black-piece" x="280" y="280">♟</text> <!-- d5 -->
    <text class="piece black-piece" x="600" y="280">♟</text> <!-- h5 -->

    <!-- File labels -->
    <text class="label" x="40" y="660">a</text>
    <text class="label" x="120" y="660">b</text>
    <text class="label" x="200" y="660">c</text>
    <text class="label" x="280" y="660">d</text>
    <text class="label" x="360" y="660">e</text>
    <text class="label" x="440" y="660">f</text>
    <text class="label" x="520" y="660">g</text>
    <text class="label" x="600" y="660">h</text>

    <!-- Rank labels -->
    <text class="label" x="-20" y="40">8</text>
    <text class="label" x="-20" y="120">7</text>
    <text class="label" x="-20" y="200">6</text>
    <text class="label" x="-20" y="280">5</text>
    <text class="label" x="-20" y="360">4</text>
    <text class="label" x="-20" y="440">3</text>
    <text class="label" x="-20" y="520">2</text>
    <text class="label" x="-20" y="600">1</text>
  </g>
</svg>

Render
Copy
Plain-Text Board Grid (White’s Perspective)

8 r n b q . b . r
7 p p p . k p p .
6 . . . . . n . .
5 . . . p . . . p
4 . . . N . P . P
3 . P . Q . . . .
2 P . P . P . P .
1 R N B . K B . R
  a b c d e f g h

Copy

(Legend: Uppercase = White pieces, Lowercase = Black pieces, . = empty square)

Definite: Verified by step-by-step move simulation and independent coordinate-to-piece mapping across all 30 remaining pieces, confirming the unique board state and the last move h2→h4.

# Qwen3.6-35B-A3B-Brainwaves-Nex-Uncensored-oQ4e-mtp

This model was quantized using [oQ](https://github.com/jundot/omlx) (oMLX v0.7.3) mixed-precision quantization.

The effective average is 4.925 bits/weight for the complete checkpoint.

## Quantization and Bit Distribution

| Storage format | Logical weights | Share of total | Tensor storage | Effective bits/weight |
|---|---|---|---|---|
| Affine 4-bit, group 64 | 33.018B | 91.838% | 17.297 GiB | 4.50 |
| Affine 8-bit, group 128 | 0.129B | 0.359% | 0.124 GiB | 8.25 |
| Affine 8-bit, group 64 | 2.328B | 6.474% | 2.303 GiB | 8.50 |
| BF16 | 0.478B | 1.329% | 0.890 GiB | 16.00 |
| **Total** | **35.952B** | **100%** | **20.614 GiB** | **4.925** |

## Storage Breakdown

| Component | Logical weights | Storage |
|---|---|---|
| Language model | 34.661B | 20.738 GB / 19.313 GiB |
| Vision | 0.447B | 0.893 GB / 0.832 GiB |
| MTP head | 0.845B | 0.503 GB / 0.468 GiB |
| **Total** | **35.952B** | **22.134 GB / 20.614 GiB** |
Catalog is the map. Apps are the tools.

Run models on your own machine, not in the cloud.

Every model page has an "Open in Abliteration" button that hands the model directly to the first-party desktop client, at the quantization your rig can actually run. No API keys, no subscription, no prompt leakage.

Get the app Browse the catalog
Open in Abliteration