Two models. Zero refusals. Full precision.

Every model below is served unquantized, uncensored at the tensor level, and available on genuinely unlimited flat-rate tiers.

2 models live 0 quantized variants 0 refusals on the A/B suite 262K max window ~100 tok/s
ModelFamilyContextSpeedModalityPrecision
GLM-5.3-Flash Uncensored
glm-5.3-flash-unquant
GLM · MoE flagship 262K ~100 tok/s text · vision · tools Unquantized View model
Qwen 3.8 27B Uncensored
qwen38-27b-abliterated
Qwen · fast workhorse 64K → 262K ~100+ tok/s text · vision · tools Unquantized View model

Served on dedicated Blackwell and RTX 5090 hardware. More models join the roster as their uncensored unquantized builds earn their place.