Uncensored CyberTiel 35B-A3B Beats Opus 4.6 on Real Codebase Issues at 27% of Qwen's Time
peculiar-ragdoll · reddit · 2026-09-11
An anonymous researcher released CyberTiel 35B-A3B, an uncensored coding model claiming to cleanly beat Opus 4.6 medium on real codebase issues while taking just 27% of the time Qwen3.8-27b medium needs.
Key technical points:
- Abliteration normally damages models, but an improved imatrix baked from a curated corpus of cybersecurity and agentic software engineering makes the damage negligible under Q4 quantization
- Weights critical to coding work are preserved at higher precision; an improved chat template makes it think and respond faster
- The model outperforms its censored TielCoder base while enabling offensive security research
- The author uses it daily over Qwen3.8-27b, fixing 3 issues in the time the 27B solves one, at the cost of reduced world knowledge
GGUF and MLX weights (with and without MTP) are on Hugging Face. The post also argues censorship undermines the user-control promise of local models.
More from coding & agent
- Microsoft demos intent-classifying agents built with Copilot App and Foundry, deployable across Teams and Outlook — marlene_zw · 2026-10-26
- Looking for a wearable mic with best battery life that feeds recordings to personal agents via MCP — altryne · 2026-10-26
- The identity crisis in software engineering: multiagent systems demand a new problem set — rakyll · 2026-10-26
- Non-CS Beginners Are Shipping Code Through CI/CD Pipelines with AI Help — curious_vii · 2026-10-26
- Ruff creator Charlie Marsh: frontier models find infinite bugs in any software — charliermarsh · 2026-10-26
- Dev testing one skill across Opus 5.5 and Sol 6.1 finds instruction following 'all over the map' — lucasmeijer · 2026-10-26