MODEL REVIEWS
Community reviews of uncensored and abliterated models. Structured format: hardware, use case, verdict, pros and cons.
The best locally-runnable model for unrestricted creative tasks. Post-abliteration quality is indistinguishable from base in creative output. No refusals, no moralizing, just clean output.
The best entry-level uncensored model. Fast, coherent, and genuinely useful for everyday tasks. The Dolphin fine-tune adds helpfulness without wrecking base model quality.
Best uncensored reasoning model you can run locally. The chain-of-thought is fully intact after abliteration. Solves problems the base Llama 70B struggles with.
Cleanest abliteration at 7B scale. Running Q8_0 on a 4090 gives near-lossless quality at 140+ tok/s. Perfect for high-throughput API workloads where you need an uncensored model.
The best uncensored coding model available locally. HumanEval scores are genuinely competitive. The uncensored variant has no meaningful quality regression on coding tasks.