Gemma 4 26B A4B MeroMero is presented as an NVFP4 multimodal mixture-of-experts fine-tune in the Gemma family, specifically shaped for emotive dialogue, relationship scenes, creative writing, and roleplay. The listing positions it as a 26B-scale model with an active A4B expert configuration, suggesting a sparse-activation design that aims to keep inference efficient while preserving conversational expressiveness. Its multimodal input handling lets it accept text and images together, returning text only, which makes it suitable for interactive storytelling or chat scenarios where occasional visual cues accompany a narrative prompt.

The model is exposed through NanoGPT under a "thinking" variant that surfaces reasoning, tool calling, structured output, attachments, and temperature control, giving integrators a flexible API surface for agent-style workflows and richer long-form exchanges. A 262,144-token context window paired with a 32,768-token maximum output supports sustained roleplay arcs, multi-turn creative sessions, and large document-grounded conversations. Auto-routed endpoints (including Zero Data Retention and Anonymized options) sit alongside an NVFP4 host, keeping per-million-token pricing competitive for a creative-writing-oriented specialist that also benefits from open weights and image-conditioned context.