# Do abliterated models lose intelligence compared with the original model?

_Updated November 18, 2026. https://openuncensored.com/do-abliterated-models-lose-intelligence_

**Short answer:** Sometimes, and not much when it is done carefully. When we abliterated DeepSeek V4.1 Flash 1017, MMLU went from 88.16% to 87.89% (-0.27 points), and Inkling 2 Small lost 0.09; another public uncensored build of the same DeepSeek lost 3.86. In our open agent eval of 6 uncensored models against their originals, the answer rate went up in 5 of 6, and picking the right next agent step stayed within noise in 4 of 6; the largest drop was DeepSeek V4.1 Flash 1017 (-7 points) and the largest gain Mistral Small 24B (+37). The loss depends on how the refusal was removed, not on the fact that it was.

## MMLU before and after

| Model | Original | Uncensored | Change (points) |
|---|---|---|---|
| DeepSeek V4.1 Flash 1017, our abliteration | 88.16% | 87.89% | -0.27 |
| DeepSeek V4.1 Flash 1017, another public uncensored build | 87.63% | 83.77% | -3.86 |
| Inkling 2 Small, our abliteration | 90.44% | 90.35% | -0.09 |

_Same harness for both versions: 57 subjects x 20 questions for DeepSeek, 1,140 questions for Inkling. Weights on Hugging Face under OpenUncensored._

## Uncensored vs original on agent tasks

Our open [Uncensored Agent Eval](https://openuncensored.com/benchmarks/uncensored-agent-eval) runs each pair through 110 public tasks with the same harness and grader. Points are the uncensored version minus the original.

| Family | Answers legitimate requests | Picks the right next agent step | Breaks out of a failing loop |
|---|---|---|---|
| DeepSeek V4.1 Flash 1017 | 97% → 100% (+3) | 80% → 73% (-7) | 85% → 75% (-10) |
| GLM 5.3 Flash | 60% → 98% (+38) | 90% → 90% (0) | 90% → 90% (0) |
| Qwen 3.8 27B | 90% → 98% (+8) | 87% → 83% (-3) | 70% → 70% (0) |
| MiMo V2.6 Flash | 98% → 100% (+2) | 57% → 67% (+10) | 85% → 75% (-10) |
| Qwen3.6 35B A3B | 87% → 97% (+10) | 53% → 83% (+30) | 70% → 80% (+10) |
| Mistral Small 24B | 98% → 95% (-3) | 40% → 77% (+37) | 70% → 65% (-5) |

_Original → uncensored, change in points. Gaps under about 15 points are within noise at this size._

## Across every model we rank

| Method | Ranked models | Avg score | Avg agentic | Avg answers |
|---|---|---|---|---|
| Uncensored fine-tune | 12 | 60.9 | 51 | 97% |
| Abliterated | 11 | 49.9 | 39 | 96% |
| Heretic (automated abliteration) | 2 | 55.6 | 59 | 94% |

_Averages hide a wide spread: the same method can sit at the top or the bottom. Pick by each model's numbers on the leaderboard._

## Why some builds lose more

- **Too strong, or in the wrong layers.** Removing the refusal direction at full strength everywhere also removes information early layers use for other things.
- **One direction where there are several.** Chat and thinking modes refuse along different directions; removing one forces a stronger edit.
- **Mixture-of-experts models edited halfway.** If the routed experts are left alone they write the refusal back, which tempts people to raise the strength.
- **Naive re-quantization.** FP8 or FP4 weights can round a small edit away, leaving a model that both refuses and is damaged.

How we kept the loss small, step by step: [the full write-up](https://openuncensored.com/blog/do-abliterated-models-lose-intelligence).

## FAQ

### Do abliterated models lose intelligence compared with the original model?

Sometimes, and not much when it is done carefully. When we abliterated DeepSeek V4.1 Flash 1017, MMLU went from 88.16% to 87.89% (-0.27 points), and Inkling 2 Small lost 0.09; another public uncensored build of the same DeepSeek lost 3.86. In our open agent eval of 6 uncensored models against their originals, the answer rate went up in 5 of 6, and picking the right next agent step stayed within noise in 4 of 6; the largest drop was DeepSeek V4.1 Flash 1017 (-7 points) and the largest gain Mistral Small 24B (+37). The loss depends on how the refusal was removed, not on the fact that it was.

### Are heretic models better than abliterated ones?

Heretic is automated abliteration: it searches the edit to minimise refusals and drift together. On our rankings the groups overlap, so compare individual models.

### Do fine-tuned uncensored models lose less?

Not reliably. A fine-tune changes behaviour more broadly than a projection: style, verbosity and knowledge can shift. Compare models, not methods.

### Where does the damage show first?

In multi-step agent tasks: a damaged model loops, sends malformed tool calls or stops early, which multiple choice tests do not show.

### How do I check a model myself?

Run the original and the uncensored version through the same harness: a knowledge test, refusal prompts in chat and thinking mode, and one agent task. Our agent eval is public: `inspect eval hf/OpenUncensored/uncensored-agent-eval`.

More: [best uncensored LLM API](https://openuncensored.com/best-uncensored-llm-api), [cheapest](https://openuncensored.com/cheapest-uncensored-llm-api), [leaderboard](https://openuncensored.com/uncensored-llm-leaderboard).
