Models

pplx-decider is a 27B decision model from Perplexity, fine-tuned from Qwen3.8-27B

Fine-tuned Qwen 3.8 27B (OBLITERATUS/Qwen3.8-27B-OBLITERATED) specialized for offensive and defensive cybersecurity, penetration testing, reverse engineering, and code execution. Trained on 2,000 high-context technical samples with native 256K context, to

Plain-spoken Qwen3.8-27B fine-tune: 42% less slop, 73% fewer em dashes, same coding-agent pass rate. MTP speculative decoding built in.

RANA: an abliterated, uncensored Qwen3.8-27B with vision, thinking and tools, from Abliterant.

Qwen3.8-27B GSQ-RCO IQ3_S (11 GB) by ISTA-DASLab, packaged with Ollama's qwen3.8 renderer so it works with Claude Code and other agent tools.

Qwen3.8-27B TURBO, uncensored, sized for 16 GB GPUs: 108k context, fast thinking, works in Claude Code and Codex.

Text-only Qwen3.8 27B GGUF files for tool calling. Default: iq2_xxs. TurinTech models, built with Artemis. Measured IQ2_XXS: Mean KLD 1.391, Same top p 68.6% vs BF16 on All 250 texts, each text weighted equally; protocol rtx5090-2026-10-25.

Qwen3.8-27B tensor-level abliterated, vision tower and MTP head untouched. 0% over-refusal on XSTest, 0-6% refusal across the A/B suite, no measurable capability loss. Full mmproj vision, tool calling and thinking, 262K context.

Qwen 3.8 27B (IQ4_XS) - Optimized for 16GB VRAM

Quants from Q2 up to Q5 from Unsloth K_M and UD_K_XL and builds for 12, 16 and 24 GB

Gargantua Code Gen 2: writes, fixes and runs code. Qwen2.5-Coder 7B fine-tuned for the openTARS tools, runs locally.

Qwen3.8-27B tensor-level abliterated, vision tower and MTP head untouched. 0% over-refusal on XSTest, 0-6% refusal across the A/B suite, no measurable capability loss. Full mmproj vision, tool calling and thinking, 262K context.

HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF

ThoxOS host-tier kernel model, THOX full-parameter fine-tune of Qwen3.8-27B. GATE-PASSED final-answer scope 41/49 (raw-completion verdict HELD, on card). Open weights. Q4_K_M GGUF 16.5 GB. Apache-2.0.

Abliterated (uncensored) version of **Qwen/Qwen3.6-27B**, refusal behavior reduced via targeted weight modification with the Heretic library, while preserving coherence.

Qwen 3.8 27B, Distilled against Opus 5.5, GPT 6 Astra, and Grok 4.7

Ternary Bonsai 2 27B (Qwen 3.8 27B) for xOllama

Text + Vision + MTP Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-Q5_NVFP4

Qwen2.5-Coder-7B fine-tuned for uncensored code generation. Made by RACER IS OP.