Tool
Compare uncensored LLMs side by side
Pick two to four models from the catalogue. Values come from the same sources as the catalogue, and a value no source states is shown as Not stated, not left blank or set to zero.
| Characteristic | Qwen3.8-27B | Huihui Qwen3.8-27B abliterated | Qwen3.8-27B Uncensored HauhauCS Aggressive (GGUF) |
|---|---|---|---|
| Typediffers | Base model | Abliteration | Not disclosed |
| Creatordiffers | Qwen Team (Alibaba) | huihui-ai | HauhauCS |
| Derived fromdiffers | Not applicable | Qwen/Qwen3.8-27B | Qwen/Qwen3.8-27B |
| Input | Text, Image, Video | Text, Image, Video | Text, Image, Video |
| Architecture | Dense | Dense | Dense |
| Total parametersdiffers | 27B | 27.8B | 27B |
| Active per token | Not applicable | Not applicable | Not applicable |
| Configured context | 262,144 | 262,144 | 262,144 |
| Stated contextdiffers | 262,144 natively, extensible up to 1,000,000 tokens | Not stated | Not stated |
| License tag | Apache 2.0 | Apache 2.0 | Apache 2.0 |
| Download | Open download | Open download | Open download |
| Formatsdiffers | Safetensors, FP8 | Safetensors, GGUF | GGUF |
| Creator claimsdiffers | 1 on the card | 1 on the card | 2 on the card |
| Hosted (this exact repo)differs | Featherless, OpenRouter, Venice | Featherless | None listed |
| Lowest listed output ratediffers | $1.875 out, $0.15 in per 1M (OpenRouter) | $12.00 out, $1.60 in per 1M (Featherless) | No rate for this exact repo |
| Source checked | 14 November 2026 | 14 November 2026 | 14 November 2026 |
| Details | Profile and sources | Profile and sources | Profile and sources |
The address bar keeps your selection after the # sign, so you can share this comparison. Shared comparisons are not separate pages and are not indexed.
Comparing a variant with its base
The most useful comparison is often a refusal-reduced variant against its own base model. Architecture, parameters and context should match, because abliteration edits weights without changing their shape. If they do not match, the repo may be a different build, a quantization, or carry metadata the creator did not update. Differences in license tags are worth a close look: a variant's tag does not change the base model's license.
What this table cannot tell you is how the models behave. No row here is a benchmark. For that, run your own tasks against each model, as described in how to evaluate an LLM for coding, or read our benchmark methodology.