The XField
journal.
How abliterated models work, how to set them up in the tools you use, and what's new.
LaunchNov 4, 2026DeepSeek V4.1 Flash.Now uncensored.
DeepSeek V4.1 Flash 1017 Uncensored is liveDeepSeek V4.1 Flash 1017 Uncensored is live, the newest model we serve: 1M context, 64K output, thinking on at high effort. Built for coding agents and long agent runs.GuideNov 4, 2026The agentkeeps going.
Uncensored models for coding agents: how to pick oneWhat a coding agent needs from a model, which uncensored model fits each job, why cached input sets the bill, and how to point Claude Code, Codex or OpenCode at it.ComparisonNov 3, 2026A router can'tchange the weights.
Routing doesn't remove refusalsA model router forwards your request to a model. If the weights refuse, the router can't change that. Here's what actually gets you answers, and what to check.EngineeringNov 1, 2026Same prefix.Smaller bill.
Prompt caching for long agent runsWhy agents resend their whole context every turn, how prefix caching reuses it, what silently breaks the cache, and how to read the cached-token fields in usage.GuideOct 31, 2026Break itbefore they do.
Red-team your own AI agent before someone else doesUse an uncensored model to plan and write attacks on your own LLM agent, then run them in a Python harness that scores tool calls and catches regressions.EngineeringOct 30, 2026One tokenat a time.
Build a chat interface that streamsStream chat completions from XField into a browser: a small server proxy, a reader loop, reasoning shown apart, and errors, retries and cancellation handled.GuideOct 29, 2026Read the sample.Write the rule.
Malware triage with uncensored models: from sample to detection ruleA SOC guide to triaging malware with uncensored models: decode a PowerShell downloader, pull IOCs, map ATT&CK, and draft Sigma and YARA rules you then test.LaunchOct 28, 2026Two GLMs.No refusals.
GLM 5.3 and GLM 5.3 Flash Uncensored are liveTwo uncensored GLM 5.3 models are now on XField: the full model, with the longest output we serve, and Flash, a lighter GLM for batch work.EngineeringOct 26, 2026Five thousandbad examples.
Building a scam detection dataset with GLM 5.3 FlashA working pipeline for synthetic smishing data: schema, diversity controls, batched generation, dedup, an audit pass, human review, and honest recall numbers.EngineeringOct 24, 2026One direction.Projected out.
How abliteration worksHow abliteration finds the one direction that carries refusal, removes it from the weights, and how to tell whether the edit worked or quietly broke something.LaunchOct 22, 2026Small model.Reads screenshots.
Qwen 3.8 27B Uncensored is liveQwen 3.8 27B Uncensored is now on XField: a compact model with 128K context, optional thinking and image input, for quick answers and drafts.GuideOct 21, 2026Same API.No refusals.
What is an uncensored LLM API?A plain-language guide to uncensored LLM APIs: what uncensored and abliterated mean, where refusals come from, what stays the same, and how to judge a provider.GuideOct 19, 2026Zero retention.Not zero records.
Zero retention doesn't mean zero recordsExactly what XField keeps for each API request, what it never stores, why the record exists, and the questions to ask any provider about retention.Keep your stack.
Lose the refusals.
Tell us what you are building. We reply with an invitation, a key and pricing for your volume.