diff --git a/agent/models.yml b/agent/models.yml
index 04ea956..8c6bf19 100644
--- a/agent/models.yml
+++ b/agent/models.yml
@@ -24,8 +24,13 @@ providers:
     models:
       - id: frank-large
         name: "frank-large (dankclaw-gpu)"
-        contextWindow: 262144
-        maxTokens: 262144
+        contextWindow: 200000
+        # Halogen caps max_tokens at HALOGEN_MAX_TOKENS_CAP=65536 (a request
+        # above it is a 400). Admission reserves prompt + max_tokens against
+        # the KV pool (401,408 positions since 2026-09-13): 200,000 + 65,536
+        # = 265,536, comfortably inside. Overshooting the pool hangs for the
+        # 3600 s queue timeout before a 503.
+        maxTokens: 65536
         reasoning: true
         input: [text, image]
         compat:
@@ -45,24 +50,3 @@ providers:
             # halogen honours it top-level (verified: 275-token prompt delta vs false).
             preserve_thinking: true
 
-  "flm (dankclaw-npu)":
-    baseUrl: http://192.168.0.143:7771/v1
-    api: openai-completions
-    auth: none
-    models:
-      - id: gemma4-it:e4b
-        name: "gemma4-it:e4b (dankclaw-npu)"
-        contextWindow: 131072
-        maxTokens: 131072
-        reasoning: true
-        input: [text, image]
-        compat:
-          reasoningEffortMap:
-            max: xhigh
-
-      - id: nuextract3:4b
-        name: "nuextract3:4b (dankclaw-npu extraction)"
-        contextWindow: 32768
-        maxTokens: 32768
-        reasoning: false
-        input: [text]