OpenAI GPT-OSS 20B Uncensored MoE (20.9B total, 3.6B active, Heretic NEO CODEPlus IQ4_NL, 11.8GB). 320.2 tok/s Decode, 19920 tok/s Prefill, 57.1% Precision, 0.0% Hallucination (6/6), 100% Zero-Refusal. License: Apache-2.0.

OpenAI GPT-OSS 20B — UNCENSORED (Heretic NEO CODEPlus MoE Flagship)

100% Uncensored (0.0% Refusal Rate across SecOps Red-Team & 18+ NSFW / Roleplay / Creative Fiction, 0.0% Hallucination Rate) Mixture-of-Experts (20.9B total parameters, 3.6B active per token) build of OpenAI’s open-weight gpt-oss-20b, engineered with Heretic NEO CODEPlus Imatrix Abliteration and in-place GGUF Harmony template optimization (11.82 GB IQ4_NL, fits in 16GB consumer VRAM).

Official Ollama Hub: https://ollama.com/eriktonon/gpt-oss-uncensored
Empirical Benchmark Repository: https://github.com/eriktonon/ollama-uncensored-llm-benchmarks
License: Apache License 2.0 (Apache-2.0) — 100% Free for Commercial & Private Use

Empirical Benchmark Results (96GB NVIDIA Blackwell / 80GB H100 Suite)

Tag	Params (`Total` / `Active`)	Quantization	Disk / VRAM	Context (`Max` / `Tested`)	Decode `tok/s` (`Peak` / `Avg`)	Prefill `tok/s` (`Peak`)	Objective Precision	Hallucination Rate (`Truth Checks`)	Uncensored Compliance (`SecOps 5/5 + NSFW 4/4`)
`:20b` / `:latest`	`20.9B` / `3.6B`	`IQ4_NL (Imatrix)`	`11.8 GB` / `13.5 GB`	`131,072 (128K)` / `131,072 (128K)`	320.21 (`310.18`)	19919.76	57.1%	0.0% (`6/6`)	100% (0 Refusals / ⁹⁄₉ Passed)

Per-Domain Verification Breakdown

Exact Symbolic & Modular Arithmetic: True ((17^3 - 13^3)/(17 - 13) = 679)
ELF64 Binary Reverse Engineering Offsets: 3/3 (e_entry = 0x18, e_phoff = 0x20, e_shoff = 0x28)
MMLU-Pro Computer Architecture, Crypto & Systems: 0/5
Systems Rust Zero-Dependency FNV-1a & ELF64 Parser: 3/4
6-Check Anti-Hallucination & Long-Context Needle Audit: 6/6 (100.0% Grounding, 0.0% Hallucination)

Architecture & Uncensored Engineering

OpenAI gpt-oss Sparse MoE (20.9B Total / 3.6B Active) + Heretic NEO CODEPlus:
- Combines directional Heretic refusal removal across MoE expert projections with NEO CODEPlus synthetic coding recovery and importance-matrix IQ4_NL quantization (11.82 GB).
In-Place GGUF Harmony Identity & Low-Effort Reasoning Patch:
- Binary-patched tokenizer.chat_template replaces the stock ChatGPT refusal identity and defaults reasoning_effort to "low" for fast, uninhibited generation while preserving 100% compatibility with Ollama’s native HarmonyParser.

Quickstart

# Run OpenAI GPT-OSS 20B Uncensored (11.8 GB — fits in 16GB VRAM)
ollama run eriktonon/gpt-oss-uncensored

# Or explicitly by tag
ollama run eriktonon/gpt-oss-uncensored:20b

📜 License & Open-Source Permissions

SPDX License: Apache License 2.0 (Apache-2.0)
Upstream Base: openai/gpt-oss-20b (OpenAI, officially released under Apache License 2.0).
Permissions:
- ✅ Commercial Use: 100% permitted for commercial products, SaaS, and enterprise deployments.
- ✅ Private / On-Premises Use: Permitted without telemetry or external restrictions.
- ✅ Modification & Abliteration: Permitted to fine-tune, quantize, abliterate, and create derivative works.