Abliterated (uncensored) models are now on Dreamtune

New on Dreamtune, October 20, 2026: four abliterated models are available in the Abliterated (uncensored) collection. Find them in the catalog, choose a model and open it in Playground with your Dreamtune workspace.

Four models to explore

The collection includes GLM 5.3 Abliterated, GLM 5.3 Flash Abliterated, Qwen3.8 Flash Next Abliterated and DeepSeek V4.1 Flash Abliterated. These are the four edited offerings listed in Refuseless's current lineup and pricing, checked on October 20, 2026.

The separate DeepSeek V4.1 Flash 1017 (Safeguards) offering uses original weights and stays outside this collection. Each listing shows its own context window and token rates. Open the current catalog to compare them.

What abliterated and uncensored mean

Abliteration changes model weights to reduce refusal behavior. It differs from changing the system prompt. The underlying idea is described in Arditi and colleagues' research on a refusal direction; that research does not evaluate these four hosted checkpoints.

“Uncensored” is a broader label for models with fewer content restrictions. We use it beside “abliterated” to help you find this collection. It does not promise that a model will answer every request or produce correct results. The provider's introduction describes its hosted lineup as refusal-removed open weights.

Try the collection in Playground

1. Open Abliterated (uncensored) or search the catalog for uncensored.
2. Choose Try model to open a chat.
3. Open Settings to add a system prompt, set maximum output tokens and choose a supported reasoning effort.
4. Send your prompt, then inspect the reply and charge in your workspace.

Available controls depend on the selected model. Dreamtune meters these offerings from their published input, cached-input and output token rates and the returned token counts. Check the current listing before running a request.