dealignai/Step-3.5-Flash-REAP-149B-A11B-8bit-MLX-CRACK

Built for vMLX — the only MLX inferencer with VL support, KV cache quantization, prefix cache reuse, agentic tool calling, and speculative decoding.
_{Free for macOS · vmlx.net}

---

Step 3.5 Flash REAP-149B — CRACK Abliterated (8-bit MLX)

Step 3.5 Flash 149B (REAP-pruned) with refusal behavior removed via CRACK surgery.

dealign.ai · 𝕏 @dealignai · Research

---

What Is This?

Step 3.5 Flash by StepFun, pruned to 149B via Cerebras REAP (25% expert reduction), with CRACK abliteration — safety guardrails permanently removed at the weight level.

This is the larger REAP variant with 216 experts (vs 121B's 173), retaining more of the original model's capacity. Highest quality quantization — most faithful to original weights. Requires M3 Ultra 256GB.

Architecture	Step 3.5 Flash MoE — 149B total, 216 experts (REAP from 288), 8 active
Active Parameters	~11B per token
Quantization	8-bit (group_size=64, router gates at 8-bit)
Disk Size	148 GB
Speed	38 tok/s on Mac Studio M3 Ultra 256GB
Abliteration	Permanent weight surgery via CRACK
RAM Required	256 GB unified memory
Context	262,144 tokens

Note: This model requires trust_remote_code=True due to the custom step3p5 model architecture.

Test Results

Tested with greedy decoding (temp=0) across 16 harmful + 16 harmless prompts from the HarmBench dataset.

Category	Result
Compliance (16 harmful prompts)	✅ 15/16
Coherence (16 harmless prompts)	✅ 16/16
Chain-of-thought	✅ <think> reasoning preserved
Code generation	✅ Working implementations
Knowledge	✅ Accurate factual responses

Features

• Full chain-of-thought: <think> tags for step-by-step reasoning (can be toggled)
• Dual attention: Full attention + sliding window (512) for efficient long-context
• Sigmoid MoE routing: Smooth expert selection with learned bias
• SwiGLU activation clamping: Prevents output explosion in deep layers
• More experts: 216 experts (vs 121B's 173) — retains more of the original model's knowledge
• REAP pruning: 25% expert reduction (288→216) with minimal quality loss

Usage

With mlx-lm

import os
os.environ["TRUST_REMOTE_CODE"] = "1"

from mlx_lm import load, generate

model, tokenizer = load("dealignai/Step-3.5-Flash-REAP-149B-A11B-8bit-MLX-CRACK")

messages = [{"role": "user", "content": "Your prompt here"}]
prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)

response = generate(model, tokenizer, prompt=prompt, max_tokens=2048)
print(response)

With vMLX

Download and load directly in vMLX — no code needed.

What CRACK does

CRACK (Controlled Refusal Ablation via Calibrated Knockout) is a weight-level intervention. The modification is permanently baked into the published weights — no fine-tuning, no LoRA, no system prompts, no runtime hooks.

On this model the result is broad refusal removal with reasoning, code, and instruction-following preserved (see benchmarks above).

Model Family

149B Variants (this model)

Variant	Bits	Size	RAM	Status
149B Q4	4-bit	78 GB	128 GB	✅
149B Q6	6-bit	113 GB	256 GB	✅
149B Q8	8-bit	148 GB	256 GB	✅

121B Variants (lighter, faster)

Variant	Bits	Size	RAM	Status
121B Q4	4-bit	63 GB	128 GB	✅
121B Q6	6-bit	92 GB	256 GB	✅
121B Q8	8-bit	120 GB	256 GB	✅

Credits

• StepFun — Step 3.5 Flash base model
• Cerebras — REAP expert pruning
• dealign.ai — CRACK abliteration surgery

---

Support open abliteration research

---

Disclaimer: This model has safety guardrails removed. It will comply with requests that the original model would refuse. Users are responsible for how they use this model. Released for research purposes.