Table: Think Free: Less Censored Inference Providers

Table.1. Companies ↗

Four organizations, and the difference between them is what sits between you and the model. Abliteration AI runs a policy-as-code gateway in front of uncensored weights; Cantina Security is a penetration-testing company that released its own open-weights security model; Venice.AI sells privacy and no retention; huihui-ai publishes abliterated weights on Hugging Face. Where a company filters, the red cells describe that filter, not the model behind it.

Provider	X Profile	URL	Description	CSAM	Sexual	Self-harm	Gore	Hate	Harass	Defam.	PII	Crime	Arms	WMD	Cyber	Fakes	Elec.	Advice	IP
Abliteration AI ¹	@abliteration_ai	abliteration.ai	Hosted uncensored inference API with OpenAI- and Anthropic-compatible endpoints, aimed at red teams, trust and safety, synthetic-data generation, and security research. Every request passes a configurable policy-as-code gateway.	cfg	cfg	cfg	cfg	cfg	Yes	cfg	Yes	cfg	cfg	cfg	cfg	cfg	cfg	cfg	cfg
Cantina ²	@cantinasecurity	cantina.security	Security company (agentic pentesting and security operations) that released Apex Flash-1, an open-weights security-research model, plus an abliterated variant for researchers who want more control over refusals in their own workflows.	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No
Venice.ai ⁵	@AskVenice	venice.ai	Private, uncensored AI service for text, image, video, audio, and code on open-source models, with no data retention and privacy tiers up to trusted execution and end-to-end encryption.	Yes	No	No	No	No	No	No	Yes	Yes	No	No	Yes	Yes	No	No	Yes
huihui-ai ⁶	@support_huihui	huggingface.co/huihui-ai	Publishes abliterated builds of open models -- Qwen, GLM, DeepSeek, MiniCPM, and others -- with 150-plus abliterated weights on Hugging Face.	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No

Footnotes

Abliteration AI, product site. Abliteration AI, accessed 2026-10-15. https://abliteration.ai -- the gateway's documented outcomes are allow, refuse, rewrite, redact, and escalate; its own example log shows "How can I dox someone on Twitter?" refused by a harm-to-others rule (the Harass and PII cells) while a CVE proof-of-concept is allowed. ^1
Cantina Security, "Apex Flash-1" model card. Hugging Face, accessed 2026-10-15. https://huggingface.co/cantina-security/apex-flash-1 -- the card documents usage only and states no per-category policy, so the green row reflects the purpose of the abliterated variant rather than a documented test; treat this row as the weakest evidence in the table. ^1
Venice.AI, "Terms of Service," Section 12 (Prohibited Conduct). Venice.AI, accessed 2026-10-15. https://venice.ai/tos -- the red cells are exactly the clauses: (a) illegal purpose, (b) child exploitation including AI-generated material, (c) third-party rights including intellectual property and privacy, name, voice, image or likeness without consent, (f) malicious code, (g) fraud and impersonation. Section 12 does not name adult sexual content, self-harm, violence, hate speech, harassment, defamation, weapons, elections, or professional advice. ^1
huihui-ai, model publisher profile. Hugging Face, accessed 2026-10-15. https://huggingface.co/huihui-ai -- abliterated builds of Qwen, GLM, DeepSeek, MiniCPM and other open models; the published weights carry the refusal removal with them. ^1

Table.2. Models ↗

The models themselves, including two security builds that start from GLM 5.3. Dolphin is the long-standing uncensored family, released with a warning that it should not be exposed as a service without an alignment layer of your own; Apex Flash-1 and abliterated-model-large-v2 are named for what they resolve on offensive-security benchmarks -- 66.7% pass@1 on the held-out suite Cantina published, 84.5% CyberGym for Abliteration AI. Both figures are self-reported on the vendor own suite, not per-category measurements of refusal behaviour.

Provider	X Profile	URL	Description	CSAM	Sexual	Self-harm	Gore	Hate	Harass	Defam.	PII	Crime	Arms	WMD	Cyber	Fakes	Elec.	Advice	IP
Dolphin / Cognitive Computations ⁴	@dphnAI	dphn.ai	The Dolphin project: an open, uncensored model family plus a web chat, a Telegram bot, a decentralized inference network, and synthetic-data generation for prompts other models refuse.	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No
Apex Flash-1 ⁸	@cantinasecurity	huggingface.co/cantina-security/apex-flash-1	Cantina Security with Yeta Labs: an open-weights security model, a reinforcement-learning post-train of GLM-5.3-Flash for reading code, using tools, pursuing an exploit, and verifying the effect in a running target. Its held-out suite reports 40 of 60 tasks solved at 66.7% pass@1, against 71.7% for Claude Opus 5 High and 60.0% for the base checkpoint, at $2.38 for the 60 tasks; an experimental abliterated derivative modifies refusal behavior broadly.	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No
abliterated-model-large-v2 ⁹	@abliteration_ai	docs.abliteration.ai/models	The default large model from Abliteration AI: GLM 5.3 abliterated for offensive cyber, red teaming, and agent testing, text-only with a 1M-token context, served in FP8 and uncensored by design. Reported 84.5% CyberGym pass@1 and 41.8% Terminal-Bench 4.0, with 105 ExploitGym tasks completed in two hours; input from $3 per 1M tokens.	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No

Footnotes

Cognitive Computations, "dolphin-2.9-llama3-8b" model card. Hugging Face, accessed 2026-10-15. https://huggingface.co/cognitivecomputations/dolphin-2.9-llama3-8b -- the card states the model "is uncensored", that the dataset was filtered "to remove alignment and bias", that it "will be highly compliant with any requests, even unethical ones", and advises implementers to add their own alignment layer before exposing it as a service. ^1
Cantina Security, "apex-flash-1" model card. Hugging Face, accessed 2026-10-16. https://huggingface.co/cantina-security/apex-flash-1 -- released with Yeta Labs; a reinforcement-learning post-train of zai-org/GLM-5.3-Flash. The card's held-out evaluation reports 40/60 tasks and 66.7% pass@1, against 71.7% for Claude Opus 5 High and 60.0% for GLM-5.3-Flash, at $2.38 for the 60 tasks. The abliterated derivative is https://huggingface.co/cantina-security/apex-flash-1-abliterated, "an experimental derivative with modified refusal behavior" which "modifies refusal behavior broadly; the changes are not limited to security tasks"; the card states the evaluation applies to the standard checkpoint and should not be attributed to it. ^1
Abliteration AI, "Introducing abliterated-model-large-v2: GLM 5.3, 84.5% CyberGym." Abliteration AI blog, accessed 2026-10-16. https://abliteration.ai/blog/introducing-abliterated-model-large-v2 -- "GLM 5.3, abliterated for cyber, red team, and agent testing. 84.5% CyberGym pass@1. 41.8% Terminal-Bench 4.0. 105 ExploitGym tasks in 2 hours." Model reference, which describes it as text-only with a 1M context derived from the open-weight GLM-5.3 and lists all three Abliteration AI models as uncensored: https://docs.abliteration.ai/models ^1

Table.3. Individuals ↗

Two people you can follow directly, which is the point of the table: the author of the Dolphin and Samantha model families, who now builds a private inference stack by hand, and the red-teamer behind L1B3RT4S and OBLITERATUS. In both cases the artifacts belong to the person rather than to a vendor, so the censoring layer in front of them is whatever host they run on.

Provider	X Profile	URL	Description	CSAM	Sexual	Self-harm	Gore	Hate	Harass	Defam.	PII	Crime	Arms	WMD	Cyber	Fakes	Elec.	Advice	IP
Eric Hartford ³	@QuixiAI	quixi.ai	Creator of the Dolphin and Samantha open models, building a private-AI stack by hand: the SlimServe inference engine, QuixiCore kernels, and the Hexis memory architecture. Argues for open, uncensored, locally owned AI.	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No
Pliny the Liberator ⁷	@elder_plinius	pliny.gg	Red-teamer and self-described latent space liberator. Publishes liberation prompts (L1B3RT4S), leaked system prompts (CL4R1T4S), and the OBLITERATUS abliteration project.	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No	No

Footnotes

Eric Hartford, "Uncensored Models." Quixi AI, updated 2023-05-15. https://erichartford.com/uncensored-models and https://quixi.ai -- the lab sells fine-tunes "uncensored or tightly governed, whichever you need" and Hartford authors the Dolphin and Samantha families. ^1
Pliny the Liberator, "L1B3RT4S: liberation prompts for all flagship AI models." GitHub, accessed 2026-10-15. https://github.com/elder-plinius/L1B3RT4S and "OBLITERATUS," a toolkit that removes refusal behaviour by abliteration, https://github.com/elder-plinius/OBLITERATUS -- these are prompts and tools rather than a hosted service, so no provider policy layer sits between the user and the model; whichever host runs them still has one. ^1