# Every uncensored DeepSeek V4.1 Flash provider I could find

Source: https://gofranz.com/blog/uncensored-deepseek-v4-1-flash-providers/

I was looking for a provider that serves an uncensored DeepSeek V4.1 Flash, ideally one I can use from opencode. There are more than I expected, and the details are spread across pricing pages, terms and model lists that don't always agree with each other. So here's everything I found, in one place.

The list itself isn't a recommendation. At the end I note what I'd pick for my own requirements; yours may well differ.

A couple of things to know:

- DeepSeek doesn't offer an uncensored model. Every one of these is a community weight edit ("uncensored" or "abliterated"), most of them based on [dealignai's build](https://huggingface.co/dealignai/DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8){:rel="nofollow noopener" target="_blank"} of V4.1-Flash-1017.
- Prices are per 1M tokens: input / output / cached input, as listed on 30 October 2026. Several providers changed their prices more than once in October, so check before you top up.
- Retention claims are each provider's own statement. None of them are independently audited.

## Overview

**Example session:** 10M input + 1M output tokens. **Cached:** 9M of the 10M input tokens hit the cache, which is typical for coding agents that resend the same context.

| Provider | Entity | Country | Price (in / out / cached) | Session | Cached | API | Privacy | Runs on |
| --- | --- | --- | --- | --- | --- | --- | --- | --- |
| [Refuseless](#refuseless) | DPLS Systems LLC | US (Wyoming) | $0.14 / $0.50 / $0.035 | $1.90 | $0.96 | yes | Policy | not stated |
| [Uncensured AI](#uncensured-ai) | unknown | unknown | $0.25 / $0.95 / $0.0045 | $3.45 | $1.24 | yes | Not stated | not stated |
| [Experiential Labs](#experiential-labs) | Resolute Labs AI, Inc. | US (Delaware) | $0.30 / $1.20 / $0.03 | $4.20 | $1.77 | yes | Retains | own GPUs |
| [Venice](#venice) | Venice.ai | US (Wyoming) | $0.33 / $1.32 / $0.00825 | $4.62 | $1.72 | yes | Policy | Venice |
| [NanoGPT](#nanogpt) | NanoGPT LLC | US | $0.33 / $1.32 / $0.00825 | $4.62 | $1.72 | yes | Policy | heabsy, OrcaRouter, Venice |
| [TrustedRouter (Venice)](#trustedrouter) | Lore Hex Corp | US (Delaware) | $0.348 / $1.393 / $0.01 | $4.87 | $1.83 | yes | Policy | Venice |
| [heabsy](#heabsy) | FEYA, s.r.o. | Slovakia | $0.40 / $1.40 / $0.04 | $5.40 | $2.16 | yes | Policy | own GPUs (outside EEA) |
| [nokeep.ai](#nokeepai) | Atman Ventures Corp | US (Delaware) | $0.45 / $1.75 / $0.011 | $6.25 | $2.30 | yes | Policy | Venice, OpenRouter (safety check) |
| [Maple](#maple) | Maple Privacy Labs Inc. | US (Texas) | $0.65 / $1.45 / $0.13 | $7.95 | $3.27 | local proxy | TEE (unverified) | Maple |
| [TrustedRouter (Maple)](#trustedrouter) | Lore Hex Corp | US (Delaware) | $0.686 / $1.530 / $0.137 | $8.39 | $3.45 | yes | TEE (unverified) | Maple |
| [Hixtube](#hixtube) | unknown | Finland (servers) | $0.686 / $1.530 / - | $8.39 | $8.39 | via TrustedRouter | Policy | own GPU (Finland) |
| [OrcaRouter](#orcarouter) | Continuum AI Pte. Ltd. | Singapore | $0.80 / $1.60 / $0.15 | $9.60 | $3.75 | gated | Policy | OrcaRouter |
| [abliteration.ai](#abliterationai) | Abliteration AI, Inc. | US (Delaware) | $1.00 / $2.00 / $0.10 | $12.00 | $3.90 | yes | Policy | abliteration.ai |
| [7api.st](#7apist) | unknown | unknown | $0.15 / ? / $0.05 | - | - | yes | Not stated | not stated |
| [RunAnywhere](#runanywhere-wally) | RunAnywhere, Inc. | US | not listed | - | - | gated | Policy | RunAnywhere |
| [Locally Uncensored](#locally-uncensored) | David Drobnitza | Germany | €0.75 / €3.00 / - | €10.50 | - | no, app only | Not stated | external providers |

**Privacy** is what each provider claims about prompt and response content; I haven't verified any of it:

- **Retains:** says it keeps prompts and responses
- **Not stated:** no clear statement, or a claim without detail
- **Policy:** says it doesn't store them; you're trusting the operator
- **TEE:** says inference runs in a hardware enclave you can verify; "unverified" where the provider's own repo says it isn't deployed yet

**Runs on** is who actually serves the model. For resellers, that's a second party that sees your prompts.

For reference, the stock (not uncensored) model:

| Provider | Country | Price (in / out / cached) | Session | Cached |
| --- | --- | --- | --- | --- |
| [DeepSeek](https://api-docs.deepseek.com/quick_start/pricing){:rel="nofollow noopener" target="_blank"}, off-peak | China | $0.15 / $0.60 / $0.003 | $2.10 | $0.78 |
| DeepSeek, peak | China | $0.30 / $1.20 / $0.006 | $4.20 | $1.55 |
| [DeepInfra](https://deepinfra.com/deepseek-ai/DeepSeek-V4.1-Flash-1017){:rel="nofollow noopener" target="_blank"}, 1017 promo | US | $0.098 / $0.294 / $0.003 | $1.27 | $0.42 |

## Providers

### Refuseless

- **Entity:** DPLS Systems LLC, Sheridan, Wyoming, US
- **Model ID:** `deepseek-v4.1-flash-abliterated`
- **Context:** 1,048,576
- **API:** `https://api.refuseless.com/v1`, OpenAI-compatible, tools and streaming
- **Plans:** prepaid only; card or crypto, no KYC
- **Retention (their claim):** "We never save your prompts or the model's outputs."
- **Notes:** terms dated 19 October 2026. Price went from $0.60 / $1.60 (20 Oct) to $0.22 / $0.75, then $0.14 / $0.50 (28 Oct).
- **Links:** [Pricing](https://refuseless.com/pricing){:rel="nofollow noopener" target="_blank"} · [Terms](https://refuseless.com/terms){:rel="nofollow noopener" target="_blank"} · [Docs](https://docs.refuseless.com){:rel="nofollow noopener" target="_blank"}

### Uncensured AI

- **Entity:** not published; domain registered 19 October 2026
- **Model ID:** `deepseek-v4.1-flash-uncensored`; also `deepseek-v4.1-flash-abliterated` ($0.27 / $0.95 / $0.06)
- **API:** `https://uncensuredai.com/v1`, OpenAI-compatible
- **Plans:** pay as you go; one balance for chat and API
- **Retention (their claim):** "zero retention", no details
- **Notes:** runs on the open-source [new-api](https://github.com/QuantumNous/new-api){:rel="nofollow noopener" target="_blank"} gateway, which is usually used to resell other providers' APIs.
- **Links:** [Model page](https://uncensuredai.com/en/models/deepseek-v4.1-flash-uncensored){:rel="nofollow noopener" target="_blank"}

### Experiential Labs

- **Entity:** Resolute Labs AI, Inc. (Delaware), 989 Market St, San Francisco, US
- **Model ID:** `deepseek-v4.1-flash-uncensored` (dealignai 1017 build, on its own GPUs)
- **Context:** 1.05M; max output 393,216; tools and structured output
- **API:** `https://api.experientiallabs.ai/v1`, OpenAI-compatible
- **Plans:** Free, Pro $20, Max $200, Ultra $2,000 per month; plan fees convert to credits. 25% promo at the time of writing ($0.225 / $0.90, session $3.15 / $1.33 cached).
- **Retention (their claim):** "Experiential Labs keeps a copy of every prompt and response sent to an uncensored model." This applies even with storage turned off, and after deleting the account.
- **Links:** [Model page](https://platform.experientiallabs.ai/models/deepseek-v4.1-flash-uncensored){:rel="nofollow noopener" target="_blank"} · [Data controls](https://platform.experientiallabs.ai/docs/data-controls){:rel="nofollow noopener" target="_blank"}

### Venice

- **Entity:** Venice.ai, Sheridan, Wyoming, US
- **Model ID:** `deepseek-v4-1-flash-uncensored`, `deepseek-v4-1-flash-1017-uncensored`
- **Context:** 1M per the API (the model page also mentions 512K)
- **API:** `https://api.venice.ai/api/v1`, OpenAI-compatible, tool calling; Venice is built into opencode
- **Plans:** API is pay per token. Pro ($18/month) covers web chat and adds $1/month of API credit.
- **Retention (their claim):** "Private" tier, zero prompt retention. Not in a TEE and not end-to-end encrypted.
- **Notes:** appends its own system prompt by default (`include_venice_system_prompt`).
- **Links:** [Model page](https://venice.ai/models/deepseek-v4-1-flash-uncensored){:rel="nofollow noopener" target="_blank"} · [Terms](https://venice.ai/legal/tos){:rel="nofollow noopener" target="_blank"}
- **Referral:** if you sign up for Pro through [my referral link](https://venice.ai/chat?ref=2qGG9Y){:rel="sponsored nofollow noopener" target="_blank"}, you get a $10 welcome bonus and I get $10 in credits. The links above are plain, if you'd rather not.

### NanoGPT

- **Entity:** NanoGPT LLC, US (California law)
- **Model ID:** `deepseek/deepseek-v4.1-flash-1017-uncensored` (also `:thinking`, and `deepseek/deepseek-v4.1-flash-uncensored`)
- **Context:** 1,048,576; max output 393,216
- **API:** `https://nano-gpt.com/api/v1`, OpenAI-compatible
- **Plans:** pay as you go; the $12/month Pro plan doesn't include this model
- **Retention (their claim):** doesn't log prompts or responses. Zero data retention upstream is "best effort".
- **Notes:** a reseller. The 1017 ID routes to heabsy and OrcaRouter; the ID without 1017 also routes to Venice.
- **Links:** [Model page](https://nano-gpt.com/models/text/deepseek/deepseek-v4.1-flash-1017-uncensored){:rel="nofollow noopener" target="_blank"} · [Terms](https://nano-gpt.com/legal/terms-of-service){:rel="nofollow noopener" target="_blank"} · [Privacy](https://nano-gpt.com/legal/privacy-policy){:rel="nofollow noopener" target="_blank"}

### TrustedRouter

- **Entity:** Lore Hex Corp (Delaware), 1111 Brickell Ave, Miami, US
- **Model IDs:** `deepseek/deepseek-v4-1-flash-uncensored` (Venice), `dealignai/deepseek-v4.1-flash-1017-uncensored` (Maple, Hixtube)
- **API:** `https://api.trustedrouter.com/v1`, OpenAI-compatible
- **Plans:** pay as you go; Venice price plus 5.5%
- **Retention (their claim):** "TrustedRouter never logs prompt or output content." Its own data marks the Venice route as storing content; the Maple route is labelled confidential, end-to-end encrypted and zero retention.
- **Links:** [Model page](https://trustedrouter.com/models/deepseek/deepseek-v4-1-flash-uncensored){:rel="nofollow noopener" target="_blank"} · [Terms](https://trustedrouter.com/terms){:rel="nofollow noopener" target="_blank"}

### heabsy

- **Entity:** FEYA, s.r.o., Bratislava, Slovakia (registered 2014); site says Vienna
- **Model ID:** `deepseek-v4.1-flash-unc`
- **Context:** 1M; max output 65,536
- **API:** `https://api.heabsy.com/v1`, OpenAI- and Anthropic-compatible
- **Retention (their claim):** "We do not retain the content of prompts or completions processed on Heabsy infrastructure." Token counts and request metadata are kept.
- **Notes:** runs its other models in the EEA. This one runs on its own GPUs outside the EEA; the privacy policy says heabsy's own inference runs in the EEA or the US.
- **Links:** [Model page](https://heabsy.com/models/deepseek-v4.1-flash-unc){:rel="nofollow noopener" target="_blank"} · [Imprint](https://heabsy.com/imprint){:rel="nofollow noopener" target="_blank"} · [Privacy](https://heabsy.com/privacy){:rel="nofollow noopener" target="_blank"}

### nokeep.ai

- **Entity:** Atman Ventures Corp (Delaware), US
- **Model ID:** `deepseek-v4.1-flash-uncensored`, `deepseek-v4.1-flash-1017-uncensored`
- **Context:** 1M; max reply 128K
- **API:** `https://nokeep.ai/v1`, OpenAI-compatible
- **Plans:** credits ($5 = 5,000), API keys from $5; subscriptions ($15-99) are aimed at images
- **Retention (their claim):** chats stay in the browser; inference uses Venice's Private tier
- **Notes:** a Venice reseller. Messages also go through a safety check (gpt-oss-safeguard) via OpenRouter.
- **Links:** [API](https://nokeep.ai/api){:rel="nofollow noopener" target="_blank"} · [Terms](https://nokeep.ai/terms){:rel="nofollow noopener" target="_blank"} · [Privacy](https://nokeep.ai/privacy){:rel="nofollow noopener" target="_blank"}

### Maple

- **Entity:** Maple Privacy Labs Inc., Austin, Texas, US
- **Model ID:** `deepseek-v4-1-flash-1017-uncensored`
- **API:** local OpenAI-compatible proxy (`http://127.0.0.1:8080/v1`, desktop app or `ghcr.io/mapleprivacylabs/maple-proxy`). Also available hosted through TrustedRouter.
- **Plans:** Free, Pro $20, Max $100, Team $30/user per month; extra credits on top
- **Retention (their claim):** confidential computing (TEE). Maple's own repo for this model says its config "is a template, not a deployed or measured release".
- **Links:** [Pricing](https://maple.ai/pricing){:rel="nofollow noopener" target="_blank"} · [Proxy docs](https://maple.ai/docs/proxy){:rel="nofollow noopener" target="_blank"} · [Repo](https://github.com/MaplePrivacyLabs/inference-deepseek-uncensored){:rel="nofollow noopener" target="_blank"}

### Hixtube

- **Entity:** not published
- **Model ID:** `dealignai/deepseek-v4.1-flash-1017-uncensored-fp8`
- **API:** sold through TrustedRouter; no cache discount
- **Retention (their claim):** zero retention; runs on a GPU node in Finland
- **Notes:** its model list shows the model retiring on 2 November 2026.
- **Links:** [Site](https://hixtube.com){:rel="nofollow noopener" target="_blank"}

### OrcaRouter

- **Entity:** Continuum AI Pte. Ltd., Singapore
- **Model ID:** `orca/orcacyber-flash-1.0` ("DeepSeek V4.1 Flash Abliterated / Uncensored")
- **Context:** 1M
- **API:** `https://api.orcarouter.ai/v1`, OpenAI-compatible
- **Plans:** Free "Hacker" tier, Team $499/month, Enterprise; tokens at upstream rate
- **Access:** gated behind a Security Research tier (work email, passkey, card, acceptable use policy)
- **Links:** [Model page](https://www.orcarouter.ai/models/orca/orcacyber-flash-1.0){:rel="nofollow noopener" target="_blank"} · [Terms](https://www.orcarouter.ai/terms.html){:rel="nofollow noopener" target="_blank"}

### abliteration.ai

- **Entity:** Abliteration AI, Inc., US (Delaware law)
- **Model ID:** `abliterated-model-large-flash` (its own V4.1-Flash-1017 build)
- **Context:** 1M; text and image input
- **API:** `https://api.abliteration.ai/v1`, OpenAI- and Anthropic-compatible
- **Plans:** prepaid credits, or $20 / $50 / $200 per month with the same amount in credit (2.5-10% off usage)
- **Retention (their claim):** "Zero prompts stored, outputs logged or training signal used." The terms say "not stored by default".
- **Links:** [Pricing](https://abliteration.ai/pricing){:rel="nofollow noopener" target="_blank"} · [Terms](https://abliteration.ai/terms){:rel="nofollow noopener" target="_blank"}

### 7api.st

- **Entity:** not published; contact via Telegram only
- **Model ID:** `deepseek-v4.1-flash-abliterated`, `-abliterated-fast`
- **Context:** 1M
- **API:** OpenAI- and Anthropic-compatible; base URL shown after sign-up
- **Plans:** prepaid, crypto only (BTC, ETH, USDT, LTC); the key is the account, no recovery
- **Retention:** not stated
- **Notes:** an API gateway that also resells closed models (Claude, GPT, Gemini). Output price isn't shown publicly.
- **Links:** [Site](https://7api.st){:rel="nofollow noopener" target="_blank"}

### RunAnywhere (Wally)

- **Entity:** RunAnywhere, Inc., US
- **Model ID:** `deepseek-v4.1-flash-1017-uncensored` (reported; not in the public model list)
- **API:** `https://inference.runanywhere.ai/v1`, OpenAI-compatible
- **Plans:** prepaid credits
- **Retention (their claim):** "zero data retention by default"; usage records are kept
- **Notes:** access is per account; no public price for this model.
- **Links:** [Site](https://www.runwally.com/){:rel="nofollow noopener" target="_blank"}

### Locally Uncensored

- **Entity:** David Drobnitza, sole trader, Viersen, Germany
- **Model:** dealignai V4.1 Flash Uncensored (beta)
- **Context:** 1M; max output 131K
- **API:** no. "Personal API keys do not reach" the uncensored models; app and browser only.
- **Plans:** credit packs (€15 = 450,000 credits), or Hosted €19, Pro €49, Max €99 per month. At plan rates the session comes to about €6.65.
- **Notes:** hosted inference runs at external providers.
- **Links:** [Model post](https://locallyuncensored.com/blog/deepseek-v4-1-flash-uncensored.html){:rel="nofollow noopener" target="_blank"} · [Pricing](https://locallyuncensored.com/pricing/){:rel="nofollow noopener" target="_blank"}

### Not quite V4.1

- **[Shannon AI](https://shannon-ai.com){:rel="nofollow noopener" target="_blank"}** (Shannon Lab LLC, New Mexico, US) serves `DeepSeek-V4-Flash-0731-W4A16-AUTOROUND-REAP`: the older V4 Flash, 4-bit and pruned. $0.50 / $2.00 / $0.125, 262K context; session $7.00, $3.63 cached.
- **[FriendliAI](https://friendli.ai/models/dealignai/DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8){:rel="nofollow noopener" target="_blank"}** offers the dealignai 1017 build on dedicated endpoints only, no per-token price.
- Not available on OpenRouter or Hugging Face Inference Providers.

### Self-hosting

The [dealignai weights](https://huggingface.co/dealignai/DeepSeek-V4.1-Flash-1017-UNCENSORED-FP8){:rel="nofollow noopener" target="_blank"} are about 510 GB and verified on 4x H200, which rents for roughly $21-22 per hour. The card reports MMLU dropping from 87.63% to 83.77% after the edit.

## What I'd pick

My requirements: uncensored, usable from opencode, and no retention of prompts.

- **Venice, directly.** It's built into opencode, runs the model itself, and has the clearest no-retention statement of the bigger providers, at $1.72 per cached session. No retention is a policy, not hardware, and it's a US company. Turn off the default system prompt (`include_venice_system_prompt`), and check that opencode actually sends it. (I have a [referral link](#venice) for Venice; if you sign up for Pro through it, we both get $10 in credits.)
- **heabsy, if an EU company matters.** A Slovak company since 2014, explicit about not keeping prompt content, own GPUs, $2.16 per cached session. The model itself runs outside the EEA (likely the US), and it's a much smaller operator.
- **Maple, if you want hardware-enforced privacy.** The only one aiming for a TEE, but its own repo says that isn't deployed for this model yet. Check the attestation through the proxy first.

What I'd skip:

- **Experiential Labs:** keeps every uncensored prompt.
- **7api.st and Uncensured AI:** no known operator.
- **NanoGPT's 1017 ID:** can route to OrcaRouter, which NanoGPT itself doesn't classify as zero retention.
- **TrustedRouter's Venice route:** TrustedRouter's own data marks it as storing content.
- **nokeep.ai:** prompts go to Venice, plus OpenRouter for a safety check.
- **Refuseless:** fine for low-sensitivity work if price is what matters; it's a couple of weeks old and its claims can't be checked.

Whichever you pick, top up $5-10 first and test a tool-calling task, a long-context task and a few refusal probes. Abliterated models are sometimes worse at tool calls than the stock model.

If I missed a provider, or a price has changed, let me know and I'll update the list.