AnonRouter review: Venice's uncensored models at Venice's prices, an attested relay, and privacy

AnonRouter is a privacy-focused router: one OpenAI-compatible API in front of about 370 models from other providers. We list the 9 uncensored text models on it. Venice serves 7 of them at exactly Venice’s own prices, abliteration.ai serves GLM 5.3 Uncensored, and Phala AI serves Qwen 3.8 27B Uncensored. AnonRouter names the host on every model page and in every response, and you can sign up with no email at all.

This review is based on AnonRouter’s published pricing, model catalog, docs, and policies, checked on October 28, 2026. Benchmark results from our own runs will be added when they’re complete. Live numbers are on the AnonRouter provider page.

The short version

Models

Seven models come from Venice, under AnonRouter’s own names:

• GLM 4.7 Flash Heretic ($0.07 / $0.40, 200K context), an automated abliteration of GLM-4.7-Flash and the cheapest model here.
• Gemma 4 Uncensored ($0.1625 / $0.50, 256K context), a fine-tune of Gemma 4 26B.
• Gemma 4 26B A4B Uncensored ($0.19 / $0.88, 64K context), a Heretic abliteration served over end-to-end encryption.
• Mistral Uncensored 1.2 ($0.20 / $0.90, 128K context), the Dolphin fine-tune of Mistral Small 24B that Venice sells as Venice Uncensored 1.2. Mistral Uncensored Roleplay ($0.50 / $2) is its roleplay sibling.
• DeepSeek V4.1 Flash Uncensored ($0.33 / $1.32, 1M context), an abliteration of the 552B DeepSeek V4.1 Flash.
• Qwen 3.6 Plus Uncensored ($0.625 / $3.75, 1M context).

Two come from other hosts:

• GLM 5.3 Uncensored ($3 / $5, 1M context), abliteration.ai’s own GLM-5.3 abliteration, served by abliteration.ai. It’s the priciest model here.
• Qwen 3.8 27B Uncensored ($0.30 / $1.50, 262K context), served by Phala AI on a route labeled TEE.

The two Qwen models don’t say how they were uncensored, so we mark them “Unverified” until we test them. AnonRouter also lists DeepSeek V4.1 Flash 1017 Uncensored and abliteration.ai’s Abliterated Model, but neither can be called, so we leave them out. An anonrouter/uncensored router model picks among eight of these for you. Search them all on the Models page.

Pricing

AnonRouter charges each provider’s per-token rate with no per-model markup. We checked the seven Venice models against Venice’s live API, and every price matches exactly. GLM 5.3 Uncensored costs the same $3 / $5 as abliteration.ai’s own list price.

A typical short chat (1,000 tokens in, 500 out) costs:

• about $0.27 per 1,000 chats on GLM 4.7 Flash Heretic, the same as at Venice
• about $0.65 per 1,000 chats on Mistral Uncensored 1.2
• about $5.50 per 1,000 chats on GLM 5.3 Uncensored

The cost is at the top-up. Each credit purchase carries a 5% service fee with a $0.75 minimum, a launch promotion; the regular rate is 8% with a $1 minimum. The minimum hits small top-ups hardest: $5 of credit costs $5.75, an effective fee of 15%. $10 costs $10.75 (7.5%), and from $15 up you pay the plain 5%. Credits come in whole dollars from $5 to $1,000, last 365 days, and purchases are final. At that size, the fee is in line with OpenRouter’s 5.5% card fee; see how resellers mark up uncensored models for the rest of the market.

Several models also have cheaper fallback routes from Surplus Intelligence and Cheaper Inference, 15% to 35% below Venice. Those routes are labeled Anonymous: the host makes no promise about keeping your prompt. There’s a free tier, but none of its models are uncensored, and free requests also run on Anonymous routes.

Privacy

AnonRouter scores 80 / 100 on our privacy rubric, the highest in the index. OpenRouter is 2nd at 65.

AnonRouter’s own layer is unusually well built. Its relay runs in an Intel TDX confidential VM that anyone can attest, and the code that touches your prompt is public at github.com/anonrouter/confidential-content-plane. In the default ticket flow, the part that sees your prompt never learns your account. The plain OpenAI-compatible key mode, on by default for new keys, does join the two in memory inside the enclave.

The big caveat: the model host still reads your prompt on most routes. The score covers AnonRouter, not the company running the model. Five of the Venice models and GLM 5.3 Uncensored are labeled Private: the host promises not to keep the content, but it still sees it. On the Venice routes, Venice’s own promise applies, and Venice scores 55 on its own. Qwen 3.6 Plus Uncensored is labeled Anonymous, so its host makes no retention promise at all. That’s why there are no technical-guarantee points.

Two routes go further, with limits:

• Gemma 4 26B A4B Uncensored has an E2EE route via Venice. Your client encrypts the prompt so only a verified enclave can decrypt it. AnonRouter’s own notes for this route say that enclave is a verified Venice gateway, which decrypts and forwards the prompt inside Venice. E2EE requests never fall back to another route.
• Qwen 3.8 27B Uncensored runs on a Phala AI route labeled TEE, and Gemma 4 26B A4B has a Phala TEE route too ($0.15 / $0.70). AnonRouter’s docs contradict this. Its TEE-verification page classifies Phala AI as private, not confidential: a gateway that decrypts before forwarding, with no route attestation. The live catalog labels both Phala routes TEE anyway.

The routing controls are the practical strength. You can set a minimum privacy level per request or per workspace, pin or exclude providers, and cap the price. Fallback reaches up to 3 providers and never drops below the privacy level you allow. For how this compares with the rest of the market, see how private uncensored AI really is.

Access and payment

• OpenAI-compatible API and a web chat. Both need an account.
• Anonymous signup gives you a generated username, a password and a 12-word recovery phrase, with no email. Lose the phrase and you may lose the account. Email, Google and GitHub signup also work.
• Card, through a third-party processor, or Bitcoin on-chain. The pricing page and terms also list Lightning, which the live checkout doesn’t offer. The credits docs say crypto isn’t enabled in production, while the live billing catalog shows Bitcoin enabled.
• Run by Sunday Labs LLC of Sheridan, Wyoming. The terms set Wyoming law and binding individual AAA arbitration, with a class-action waiver and a 30-day opt-out. Users must be 18 or older.

How it compares

• Straight from the source: Venice (55) sells the same seven models at the same prices with one fewer company in the path, but needs an account.
• Bigger fine-tune catalog: OpenRouter (65) lists 20 uncensored models, including Hermes and Euryale, but no abliterated ones, and needs an account.
• Also an enclave relay: PayPerQ (45) decrypts prompts only inside an open-source AWS Nitro Enclave, needs no account, and takes Lightning and Monero. It charges about 5.5% over upstream.
• Biggest catalog: NanoGPT (35) lists 137 models, takes crypto, and needs no signup.

Verdict

AnonRouter earns its first place on the things a router controls. It doesn’t store prompts, it names the host on every request, its relay is open source and attestable, and you can sign up without an email. The prices are Venice’s own, so the only extra cost is the top-up fee. Buy at least $15 at a time to keep it at 5%.

The score is about AnonRouter, not the model host. On most routes Venice or abliteration.ai still reads your prompt under its own policy. The E2EE Gemma route ends at a Venice gateway, and AnonRouter’s docs disagree with its own catalog about the Phala TEE label. If you only want Venice’s models and don’t mind an account, Venice directly is one less party. If you want anonymous signup, per-request host control and Venice’s prices together, AnonRouter is the one to use.

Compare every AnonRouter model and price on its provider page, or see every provider in the index.