Abliterated research variant of GLM-5.3-Flash: refusal direction removed from the weights (weight orthogonalization, no adapters or runtime code), served unaligned for red-teaming and interpretability research. Same 320B/18B-active hybrid MoE architecture and 1M context as the base model; capability within measurement noise of base (MMLU 87.3 vs 86.7).
Context
1.05M
Max output
131K
Price per million tokens
$0.15in$0.030cached$0.50out
Waterfall
Quickstart
I want you to route my LLM calls for "glm-5.3-flash-uncensored" through the Experiential gateway instead of calling the provider directly. It speaks the OpenAI Chat Completions API, so this is a base-URL and key swap. Please: 1. Point the client at https://api.experientiallabs.ai/v1 as the base URL. 2. Authenticate with my Experiential API key from the EXPLABS_API_KEY environment variable. If it isn't set, stop and tell me to create one under Settings -> API Keys and export it. 3. Use the model id "glm-5.3-flash-uncensored" exactly. 4. Update every place my code builds an LLM client for this model to use that base URL and key, leaving streaming and tool-calls as they are. 5. Make one test call and show me the reply plus the token usage, so we confirm it runs on my Experiential credits. Tell me which files you changed.
Sampling
Tools & structure
Fastest
45 tok/s
Released
-
Reasoning
Streaming
Limits
Sending an unsupported field? See error reference