schema_version = 1

extends     = "LM Studio Chat"
name        = "LM Studio Chat — 16 GB (Gemma 4 12B)"
description = "Light tier. Local LM Studio coding chat on Gemma 4 12B QAT (Q4_0) — ~7 GB weights, ~10 GB resident at 8K context; comfortable on 16 GB (runs on 12 GB). Fastest to load, best for quick iterations. Native reasoning + tools + vision. Point the LM Studio (Responses API) provider URL at your host. Note: on LM Studio the context length (KV cache) is a model-load setting, not per-request — the agent's only memory lever is the model."

model           = "google/gemma-4-12b-qat"
enable_thinking = true
tags            = ["chat", "lmstudio", "responses", "local", "gemma4", "16gb"]

[body]
temperature       = 1.0
top_p             = 0.95
max_output_tokens = 16384