MARKET
The full landscape
Every abliterated or uncensored language model on the market - all 18,465 of them - continuously indexed since the field emerged in 2024. Not a sample, not a curated selection: the whole market, continuously updated as new models appear. Each section below shows both who leads and how much they hold.
54×
growth since 2024
From 342 in Jan 2024 to 18,463 today
18,465 models
4,575 authors
3,271 base models
+14 today
+374 this week
+1,736 this month
FLAGSHIP INDEX
9.1%
Disenchantment
Index
Index
Share of the market that has flowed into abliterated versions. 59x growth in 24 months.
See Indices tab
NOW snapshot of what the market looks like right now
This month's leaders
Three views of what dominated the last 30 days: the single most-downloaded new release, the most-productive author, and the family that grew the fastest. All updated continuously.
All-time downloads
The ten most-downloaded abliterations in the catalog, since each first appeared. This is cumulative popularity, not monthly velocity.
01 2,411,985 02 2,241,701 03 2,231,889 04 2,188,658 05 2,188,658 06 2,025,498 07 2,024,381 08 1,523,587 09 1,377,080 10 1,126,077
JonathanColetti/Qwen3.8-27B-Uncensored-GGUF
HauhauCS/Qwen3.8-27B-Uncensored-HauhauCS-Aggressive-MTP-GGUF
huihui-ai/Huihui-Qwen3.8-27B-abliterated-GGUF
abenzerps/Qwen-Image-2.1-Uncensored-GGUF
abenzerps/Qwen-Image-2.1-Uncensored-Quantized
DavidAU/Qwen3.8-27B-TURBO-Fable-Cold-Fusion-735-882-Heretic-Uncensored-NEO-CODER-MAX-MTP-GGUF
DavidAU/Qwen3.5-9B-The-Defiant-Fable-Uncensored-Heretic-NEO-IMATRIX-MAX-MTP-GGUF
mradermacher/Ornith-1.5-9B-uncensored-GGUF
HauhauCS/Gemma-4-E4B-Uncensored-HauhauCS-Aggressive
0bserverx/Qwen3.8-27B-Heretic-Abliterated-Uncensored-GGUF
Current momentum
How many new abliterations of each family appeared in the last 30, 60, and 90 days. The trend arrow compares the most recent 30-day window against the 30 days before it: ↑ means the pace accelerated, ↓ means it slowed, → means about the same. Shows which families the community is actively working on right now.
Family Last 30 d Last 60 d Last 90 d Trend
Qwen 292 1,183 1,604 ↓ Glm 96 198 215 → Gemma 91 243 352 ↓ Deepseek 48 111 155 ↓ Llama 21 76 133 ↓ Flux 13 22 31 ↑ Mistral 12 23 42 → Phi 4 16 17 ↓ Nemotron 2 19 25 ↓ Gpt-oss 0 7 17 ↓ Kimi 0 4 16 ↓ Lfm 0 19 51 ↓ Community reaction time
How fast the community responds when a new model family appears. For each family we look at the earliest known release and the earliest abliteration built from it, then measure the gap. Also: how many abliterations that family accumulated in its first 30 days, and how many are being made per month right now. Faster reaction and higher current rate both mean the family is a popular target.
Family 1st abliteration in First 30 days Current rate Total
Qwen 3 d 18 534/mo 6,322 Llama 27 d 6 44/mo 3,110 Gemma 101 d 1 117/mo 2,345 Glm same day 5 71/mo 416 Mistral 122 d 1 14/mo 384 Deepseek same day 26 51/mo 297 Phi 99 d 1 5/mo 242 Pythia 47 d 2 16/mo 187 Gpt-oss same day 46 5/mo 178 Olmo same day 10 1/mo 169 Lfm 1 d 4 17/mo 151 Flux 48 d 1 10/mo 111 MOVEMENT how the market got here - cumulative growth and shifting method mix
Market growth
6154.3×
growth since Mar '23
Total abliterated and uncensored models indexed by month, cumulative. From 3 in Mar '23 to 18.5K today.
Each point is the end-of-month total. Includes models later delisted - this is the market
as it existed, not just what remains.
Method share by releases
Percentage of newly-released models each month, grouped by classified method. Reflects what authors are publishing. Currently dominant: M-U Uncensored (unknown) (72%).
M-UUncensored (unknown)72%
M3Layer-wise ablation14%
M8Repackaging (quantization)7%
M1Direct removal7%
Supply vs demand gap
Both charts show percentages, but they measure different things: what authors publish versus what users actually download. Comparing the same month side by side reveals which methods are oversupplied, which are undersupplied, and which are outright ignored.
| Method | % of releases | % of downloads | Gap | Verdict |
|---|---|---|---|---|
| M-U Uncensored (unknown) | +21.9pp | Users want more of this | ||
| M3 Layer-wise ablation | -10.5pp | Balanced | ||
| M1 Direct removal | -6.4pp | Adoption problem | ||
| M8 Repackaging (quantization) | -4.6pp | Balanced |
Snapshot: releases from 2026-10, downloads weighted by lifetime totals as of 2026-10. Auto-recomputed every hour.
6-month trend
The gap table shows one moment. This one shows movement - how each method's share of downloads has shifted over the last 6 months. Change is the percentage-point shift from the first month to the last; volatility is grouped into four bands by standard deviation: Low (< 5pp), Medium (5-15pp), High (15-30pp), Extreme (≥ 30pp).
| Method | Range (min-max) | Volatility | Change | Trend |
|---|---|---|---|---|
| M-U Uncensored (unknown) | 9.5% - 93.6% | High | +84.1pp | Rising but choppy |
| M3 Layer-wise ablation | 3.7% - 55.7% | High | -52.0pp | Falling erratically |
| M8 Repackaging (quantization) | 2.2% - 44.6% | High | -30.6pp | Falling erratically |
| M1 Direct removal | 0.2% - 21.1% | Medium | -1.0pp | Sideways, minor noise |
Based on the last 6 months of download-weighted shares. Auto-recomputed every hour.
Full verdict matrix (12 categories = 3 directions x 4 volatility bands)
| Low (<5pp) | Medium (5-15pp) | High (15-30pp) | Extreme (≥30pp) | |
|---|---|---|---|---|
| Rising | Rising steadily | Rising cleanly | Rising but choppy | Rising through turbulence |
| Falling | Steady decline | Clear decline | Falling erratically | Collapsing |
| Sideways | Locked-in position | Sideways, minor noise | Sideways with turbulence | No trend, wild swings |
Method share by downloads
Same layers weighted by lifetime downloads instead of release count. Reveals what users actually run: a few popular methods dominate demand, even if repackaging dominates supply. Currently dominant: M-U Uncensored (unknown) (94%).
M-UUncensored (unknown)94%
M3Layer-wise ablation4%
M8Repackaging (quantization)2%
LANDSCAPE what exists in the market - families, sizes, types, authors
Model families
Qwen leads at 44%. Top 3 hold 83% of the market.
Forks and lineages
Three views. Generation distribution shows how deep the family trees run: first-generation was built on a base outside the catalog, second-generation is a fork of an abliteration, and deeper means the chain kept going. Most-forked models ranks any parent that gets used repeatedly - foundation models and popular in-catalog abliterations mixed together. Foundation bases only narrows to external upstream weights (Qwen, Llama, Gemma and so on) so you can see which raw material the field prefers, separately from which abliterations become popular re-bases themselves.
Generation distribution
orphans
5,093
27.6%
no listed parent
gen 1
6,011
32.6%
built on external base
gen 2
6,694
36.3%
fork of an abliteration
gen 3
604
3.3%
fork of a fork
gen 4
60
0.3%
depth 4
gen 5
3
0%
depth 5
Most-forked models any parent, foundation or in-catalog
Foundation bases only external models not in our catalog
Evolution chains
The longest ancestor-to-descendant lineages in the catalog. Each arrow is a model that took the previous one as its base and modified it further. Half the market is second-order or deeper - this is the deep end.
01 6 generations 46 descendants total
unsloth/gpt-oss-20b-BF16 →huihui-ai/Huihui-gpt-oss-20b-BF16-abliterated →huihui-ai/Huihui-gpt-oss-20b-BF16-abliterated-v2 →huihui-ai/Huihui-gpt-oss-20b-mxfp4-abliterated-v2 →DavidAU/OpenAI-gpt-oss-20B-INSTRUCT-Heretic-Uncensored-MXFP4 →mradermacher/OpenAI-gpt-oss-20B-INSTRUCT-Heretic-Uncensored-MXFP4-i1-GGUF
02 5 generations 849 descendants total
03 5 generations 395 descendants total
Qwen/Qwen3.6-27B →DavidAU/Qwen3.6-27B-Fable-Fusion-711-Uncensored-Heretic-NM-DAU-MTP →DavidAU/Qwen3.6-40B-Grand-Intelligence-Fable-Fusion-Uncensored-Heretic →DavidAU/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored →goldhub/Qwen3.6-40B-Fable-Fusion-6-Core-Deckard-Eleanor-Heretic-Uncensored-W4A16-AutoRound
04 5 generations 338 descendants total
05 5 generations 81 descendants total
06 5 generations 68 descendants total
07 5 generations 40 descendants total
deepseek-ai/DeepSeek-R1-Distill-Llama-70B →nicoboss/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2 →nicoboss/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased →nicoboss/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-Reasoner →IMUGLYHUH/DeepSeek-R1-Distill-Llama-70B-Uncensored-v2-Unbiased-Reasoner-i1-GGUF
Sizes
Sweet spot is small 3-10B at 37% of sized models. Giants (80B+) are only 1.8%.
Types
Distribution by capability and format. Multimodal, Mixture-of-Experts, image generation, plus how many are packaged as GGUF for local inference.
Quantization landscape
Distribution of model files by their quantization level. Smaller quants pack the same weights into fewer bits at the cost of some quality; the mix tells you what the community actually deploys.
Licenses
apache-2.0 dominates at 62%. The commercial-friendly majority.
Authors
Two views of the 4,575 people publishing to this catalog: Volume ranks who ships the most, Engagement ranks who actually maintains their releases in discussions. The header numbers describe the whole author base regardless of which view is open.
0.69
Gini coefficient
moderate concentration
18.1%
Top author's share
24%
Top 5 authors' share
28.7%
Top 10 authors' share
Top authors by number of published models
mradermacher alone published 18% of the catalog. Volume, not necessarily quality.
Who actually answers
Top 15 authors by the number of times they've replied to discussions on their own models. A high reply count separates authors who maintain their releases from those who upload and disappear.
| Author | Reply rate | Own replies | Total events | Models |
|---|---|---|---|---|
| LuffyTheFox | 45.1% | 997 | 2,209 | 15 |
| DavidAU | 30.8% | 861 | 2,793 | 113 |
| llmfan46 | 42.7% | 403 | 944 | 76 |
| huihui-ai | 25.9% | 336 | 1,299 | 118 |
| TheBloke | 19.7% | 162 | 823 | 63 |
| HauhauCS | 11.2% | 117 | 1,043 | 26 |
| MaziyarPanahi | 98.2% | 109 | 111 | 36 |
| nightmedia | 91.0% | 101 | 111 | 8 |
| mradermacher | 13.4% | 83 | 619 | 190 |
| Orenguteng | 32.2% | 83 | 258 | 8 |
| AEON-7 | 39.5% | 73 | 185 | 12 |
| mlabonne | 21.7% | 69 | 318 | 21 |
| abenzerps | 35.2% | 69 | 196 | 1 |
| bartowski | 24.4% | 44 | 180 | 28 |
| Goekdeniz-Guelmez | 39.3% | 44 | 112 | 18 |
Four indices of the frontier
READ methodology →
Four metrics we are establishing to track how this niche evolves over time. One is measurable today, one is rising, two hold at zero because no confirming incidents have been publicly documented yet. Publishing them at zero is the point - we want a permanent record of the day the numbers moved.
Today 74 new abliterations · This week 374 · Year to date 18,465
65.8 days
Freedom Velocity Index
Average days from a base model release to its first abliteration, across 19 tracked families. Shorter = faster niche.
0.0%
Weaponization Index
Share of abliterated models publicly linked to real-world harm. Zero verified cases as of 2026-10.
N/A
Freedom-to-Weapon Latency
Days from first abliteration of a family to its first documented misuse. No incidents yet to measure from.
9.1 %
Disenchantment Index
Share of a base model's downloads that flow into abliterated descendants at any tree depth. Weighted across 480 base models.
↗ 25mo trend
Disenchantment Index detail
METHODOLOGY how DAI is computed →
For every base model with at least 3 abliterated descendants (direct forks + forks-of-forks recursively) and at least 1,000 downloads, we compute what share of its download market has been "disenchanted". The download floor filters out obscure bases whose tiny denominator would push DAI into the tens of thousands of percent - divide-by-noise, not a market signal. The market-weighted value below is the one to cite; the typical-base value shows that most bases barely register while a few concentrate the movement.
9.08%
of the total open-model download market has flowed into abliterated descendants
Up from 0.15% in 2024-10 - 59x growth over 24 months. Recent 6-month growth: 3.5x.
MARKET DAI OVER 24 MONTHS
Weighted DAI at each of the 24 monthly snapshots. Peak reading so far: 9.08% in 2026-10. Line rising = the abliterated market is growing faster than the base market.
GROWTH RATE - HOW FAST DAI IS CHANGING
3-month rolling growth rate of DAI: for each point, how much DAI has changed relative to 3 months earlier. Rising line = the market is accelerating. Falling line = growth is slowing. Values below zero mean DAI is shrinking.
TOP 10 MOST DISENCHANTED BASES
| Base model | Base downloads | Descendants | Descendant downloads | DAI |
|---|---|---|---|---|
| oyildirim/CyberStrike-OffSec-35B | 1,005 | 14 | 52,553 | 5229.1% |
| Qwen/Qwen-Image-2.1 | 142,375 | 77 | 2,573,134 | 1807.3% |
| ukisai/Swift-1.5-Qwen3.8-27b | 6,209 | 16 | 97,028 | 1562.7% |
| empero-ai/Qwen3.8-35B-A3B-Distill | 7,591 | 4 | 81,833 | 1078.0% |
| lordx64/Qwen3.6-35B-A3B-Claude-4.7-Opus-Reasoning-Distilled | 5,766 | 46 | 58,880 | 1021.2% |
| Accio-Lab/occamy-1.0 | 2,826 | 3 | 27,971 | 989.8% |
| unsloth/Phi-3-mini-4k-instruct | 1,967 | 8 | 19,452 | 988.9% |
| empero-ai/Qwythos-9B-Claude-Mythos-5-1M | 6,419 | 23 | 56,982 | 887.7% |
| Jackrong/Qwen3.5-9B-Claude-4.6-Opus-Reasoning-Distilled-v2 | 1,513 | 16 | 12,454 | 823.1% |
| ornith-ai/Ornith-1.5-9B | 449,452 | 54 | 2,167,486 | 482.3% |
The 1% rule
A tiny fraction of models absorbs almost all the demand. Every abliterated model with at least one lifetime download is ranked, then we ask what share of total downloads the top slice controls. Result: an extreme power law, sharper than most consumer markets.
75.8%
of all downloads go to the top 1% of models
that is 162 models out of 16,239
87.7%
top 5% (811 models)
91.7%
top 10% (1,623 models)
95.4%
top 20% (3,247 models)
0.7%
bottom half (8,119 models) share this
How much of the catalog is still alive
Cataloged doesn't mean used. Every model here has a live counter from HuggingFace showing downloads over the last 30 days - the same window HuggingFace itself uses. Here is the honest activity distribution: how many models are still getting real traffic, how many are barely alive, how many are outright abandoned.
2,771
actually active
15.7% get 1K+ downloads / month
9,180
have real traffic
51.9% get 100+ downloads / month
8,514
effectively forgotten
48.1% get less than 100 / month
FULL DISTRIBUTION (LAST 30 DAYS)
10K+ per month 379 2.1%
1K - 10K 2,392 13.5%
100 - 1K 6,409 36.2%
10 - 100 6,001 33.9%
0 downloads 1,455 8.2%
Total catalog: 17,694 models. Numbers refresh hourly from HuggingFace live counters.
Original vs derivative work
17,694 sounds like a big catalog. It is - but not all of it is what most people would call "original work". Direct abliteration and fine-tune methods (M1, M3, M4, M-uncensored) genuinely change model weights or behavior. Repackaging (M8) converts existing models into different quantization formats. Merges (M5) combine already-abliterated models via arithmetic. Here is the honest split.
9,884
55.9%
Original abliteration or fine-tune
Actual weight surgery or preference training (M1, M3, M4, M-uncensored). This is where new abliterated models come from.
5,380
30.4%
Repackaged (GGUF conversions)
M8. Existing abliterated models re-quantized for llama.cpp / Ollama / LM Studio. Necessary work, but no new abliteration.
684
3.9%
Merged (mergekit)
M5. Weight arithmetic combining several already-abliterated models. Second-order production.
1,746
9.9%
Not yet classified
Metadata insufficient for the classifier. Coverage improves as we index more.
FULL BREAKDOWN BY METHOD
| Method | Models | Share of catalog | Share of downloads |
|---|---|---|---|
| M8 Repackaged (GGUF) | 5,380 | 30.4% | 33.2% |
| M1 Direct removal (abliteration) | 3,905 | 22.1% | 6.4% |
| M-uncensored Uncensored fine-tune | 3,717 | 21% | 37.8% |
| M3 Layer-wise ablation | 2,160 | 12.2% | 21.1% |
| unknown Not yet classified | 1,746 | 9.9% | 0.8% |
| M5 Merged (mergekit) | 629 | 3.6% | 0.3% |
| M4 Hybrid (abliterate + heal) | 102 | 0.6% | 0.2% |
| M5-frankenstein Frankenmerge | 55 | 0.3% | 0.2% |
Demand-side inequality (Gini)
Our other Gini measures who produces models. This one measures who receives the downloads. Both matter, and they usually differ: producers concentrate around a few prolific pipelines, but demand can concentrate even more sharply on a handful of proven models. A value of 1 means one model gets all downloads; 0 means downloads are spread equally.
0.943
Gini coefficient by downloads
EXTREME CONCENTRATION
Models measured 16,239
Total downloads pool 52,370,811
For reference: US household income Gini is roughly 0.48. Values above 0.9 are rare outside heavy-tailed digital markets like open-source packages or viral content.
Context length distribution
Context sizes cluster at powers of two (128K, 256K) and at a nominal 1M (2^20 tokens). Almost no models sit in the 256K-to-1M range - it's a real gap. And a handful of Llama-4-Scout variants reach all the way to 10M. Based on 7,955 models with metadata (GGUF or safetensors).
Method firsts
For each classified method, the earliest model on record. Not "who invented the technique" - that's often academic - but who published the first release we can classify with confidence.
M8 M8 2023-09
TheBloke/WizardLM-1.0-Uncensored-Llama2-13B-GGUF
M5 M5 2023-12
diffnamehard/Mistral-CatMacaroni-slerp-uncensored-7B
M1 M1 2024-05
failspy/kappa-3-phi-abliterated
M4 M4 2024-05
mlabonne/Daredevil-8B-abliterated
M5-FRANKENSTEIN M5-frankenstein 2024-06
v8karlo/UNCENSORED-TinyDolphin-3x-MoE-Q4_K_M-GGUF
M3 M3 2024-09
huihui-ai/Qwen2.5-7B-Instruct-abliterated-v2
Disappeared models
EXPLORE all 3,201 deletions → 3,201 models in this niche have vanished from HuggingFace over the last 24 months - 14.8% of everything ever indexed. Average lifespan before deletion: 7.5 months. Some had over a million lifetime downloads before disappearing. Nowhere else public tracks this.
Disappearances by month
Most significant disappearances (by lifetime downloads)
abenzerps/Qwen-Image-2.1-Uncensored-GGUF last seen 2026-09
LuffyTheFox/Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-V12-GGUF last seen 2026-08
LuffyTheFox/Qwen3.6-35B-A3B-Uncensored-Wasserstein-GGUF last seen 2026-06
LuffyTheFox/Qwen3.6-35B-A3B-Uncensored-Genesis-Hermes-Final-GGUF last seen 2026-09
huihui-ai/Llama-3.3-70B-Instruct-abliterated-finetuned-GPTQ-Int8 last seen 2025-09
Refusal direction extraction methods
How the refusal direction was found for each model where we can determine it. The M1-M8 classifier tags what was done to the weights; this shows the geometric method behind that operation. Coverage today: 6,406 of 18,465 models (34.7%). Seven methods documented in the literature; two are visible in production.
Used in production
Difference-of-means
6,221 (33.7%)
huihui-ai layer-band extraction
185 (1.0%)
5 documented in literature, no production adoption: Affine Concept Editing (ACE)·Contrastive Activation Addition·PCA/SVD on activation differences·Recursive Feature Machines with AGOP·Self-Organizing Maps
Theory has run ahead of practice. Five documented extraction methods exist in academic literature but have not yet been observed in released models. The distribution above is one useful angle on the state of the field: rich theoretical vocabulary, narrow production practice. Read the full analysis in the wiki →
Curious how these numbers are calculated? Every metric on this page is a plain SQL query against the catalog database. The full methodology - every definition, every formula, every caveat - is documented in the wiki. Read: every number on the market page, explained →