DeepSeekDeepSeek·💬 Text Generation·New

DeepSeek V4.1 Flash 1017 Abliterated🔒Private

ReasoningCodeFunction CallingWeb SearchE2EEfp8private
🧠 Try in Intelligence →Try on Venice.ai ↗
Quick reference
DeepSeek V4.1 Flash 1017 Abliterated — TLDR
  • 🔒 Served inside a Trusted Execution Environment with hardware attestation evidence
  • 🧠 Text-only reasoning and coding model with refusal behavior removed in weights
  • 📏 One-million-token context window, served in native FP8
  • 🏢 Built on DeepSeek's October 17, 2026 V4.1 Flash checkpoint
  • 🔧 Function calling and web search supported in this catalog deployment
  • 🆕 Abliteration edits a subset of layers; other components left unchanged
  • 📚 MIT license inherited from the upstream DeepSeek checkpoint
  • 💬 Latest entry in this catalog's enclave-hosted DeepSeek Flash line
💰 Pricing
$0.375 / $1.50
per 1M · input / output
📏 Context
1M tokens
📅 On Venice since
Nov 5, 2026
12 days ago
Provider

DeepSeek is a Chinese artificial intelligence company specializing in large language model development, founded in July 2023 by Liang Wenfeng. Based in Hangzhou, Zhejiang, the company is backed by High-Flyer, a prominent Chinese hedge fund also co-founded by…

Read full profile →
13 models on Venice
13 text
Since Dec 4, 2025

About this model

DeepSeek V4.1 Flash 1017 Abliterated is a modified build of DeepSeek's October 17, 2026 Flash checkpoint, served here text-only inside a Trusted Execution Environment so that enclave identity and configuration can be verified independently through hardware attestation. Abliteration refers to directional ablation of the refusal direction directly in the weights, rather than a system-prompt or runtime filter, so the model loads and runs like the upstream release.

Relative to earlier members of the family, the lineage reflects DeepSeek's own progression: DeepSeek V4 Flash and DeepSeek V4.1 Flash preceded the dated DeepSeek v4.1 Flash (1017) snapshot from which this build derives. Compared with the earlier enclave-hosted V4 Flash entry, this deployment advances to the V4.1 generation, exposes a one-million-token context window, and adds function calling and web search. Published abliterated variants apply norm-preserving edits across a subset of layers while leaving routed experts and other components byte-identical to upstream, and inherit the MIT license.

The main differentiator versus the non-enclave DeepSeek V4.1 Flash 1017 Uncensored sibling is deployment rather than model behavior: this entry adds cryptographic isolation and attestable execution. Weights are served quantized to FP8, and only text input and output are offered in this configuration.

This About section is AI-generated from public sources (Claude Opus 5), with no human editing. It may contain inaccuracies — verify critical details against the sources listed above.

Data sources: Venice API · HuggingFace · Wikipedia — enrichment updated 5d ago