diff --git a/router/router.py b/router/router.py index 66d4e06..8d1a754 100644 --- a/router/router.py +++ b/router/router.py @@ -27,7 +27,7 @@ GPU_MAX_CONCURRENT = { # Context window sizes (tokens) — used for compaction signals GPU_CONTEXT = { "qwen3.6-35B-A3B": 131072, - "qwen3.6-27B-code": 98304, + "qwen3.6-27B-code": 262144, "gemma-4-12b": 262144, # 262K context (gemma-4-12b) }