From 463590861cf43945825d7334fd2cfc79eed0bd77 Mon Sep 17 00:00:00 2001
From: adelvillar1 <adelvillar@gmail.com>
Date: Sat, 13 Jun 2026 13:04:30 -0500
Subject: [PATCH] fix(models): support glm-5.2, kimi-k2.7-code + merge
 live/curated model lists
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

Three changes to fix newly-released models not appearing in the picker
or being capped at wrong context lengths:

1. Add glm-5.2 to zai provider's static model list
2. Add kimi-k2.7-code to kimi-coding provider's static model list
3. Fix generic live-fetch path in provider_model_ids() to MERGE curated
   entries with live /v1/models results instead of replacing them —
   same pattern already used by the Anthropic path. Providers like Z.AI
   and Kimi don't list new models in /v1/models immediately even though
   inference works, causing curated entries to silently disappear.
4. Add glm-5.2 context length (1M tokens) to DEFAULT_CONTEXT_LENGTHS,
   before the generic 'glm' catch-all (202752) that was capping it at
   ~200K via substring matching.
---
 agent/model_metadata.py |  5 ++++-
 hermes_cli/models.py    | 16 +++++++++++++++-
 2 files changed, 19 insertions(+), 2 deletions(-)

diff --git a/agent/model_metadata.py b/agent/model_metadata.py
index 3a71e974fdb7..f19ccdc6a940 100644
--- a/agent/model_metadata.py
+++ b/agent/model_metadata.py
@@ -209,7 +209,10 @@ def _strip_provider_prefix(model: str) -> str:
     # https://platform.minimax.io/docs/api-reference/text-chat-openai
     "minimax-m3": 1000000,
     "minimax": 204800,
-    # GLM
+    # GLM — GLM-5.2 ships with 1M context; GLM-5 and GLM-5.1 are 200K.
+    # Substring matching is longest-first, so "glm-5.2" wins over the
+    # generic "glm" catch-all for that slug.
+    "glm-5.2": 1_000_000,
     "glm": 202752,
     # xAI Grok — xAI /v1/models does not return context_length metadata,
     # so these hardcoded fallbacks prevent Hermes from probing-down to
diff --git a/hermes_cli/models.py b/hermes_cli/models.py
index 9b473f05606a..7d3849c7df8a 100644
--- a/hermes_cli/models.py
+++ b/hermes_cli/models.py
@@ -256,6 +256,7 @@ def _xai_curated_models() -> list[str]:
         "gemini-3.5-flash",
     ],
     "zai": [
+        "glm-5.2",
         "glm-5.1",
         "glm-5",
         "glm-5v-turbo",
@@ -280,6 +281,7 @@ def _xai_curated_models() -> list[str]:
         "openai/gpt-oss-120b",
     ],
     "kimi-coding": [
+        "kimi-k2.7-code",
         "kimi-k2.6",
         "kimi-k2.5",
         "kimi-for-coding",
@@ -2330,7 +2332,19 @@ def provider_model_ids(provider: Optional[str], *, force_refresh: bool = False)
             if api_key:
                 live = _p.fetch_models(api_key=api_key)
                 if live:
-                    return live
+                    # Merge curated entries with live results so models that
+                    # are reachable for inference before they appear in
+                    # /v1/models survive the live fetch (same pattern as the
+                    # Anthropic path above). Curated go first to preserve the
+                    # recommended ordering in the /model picker.
+                    curated = list(_PROVIDER_MODELS.get(normalized, []))
+                    merged = list(curated)
+                    merged_lower = {m.lower() for m in curated}
+                    for m in live:
+                        if m.lower() not in merged_lower:
+                            merged.append(m)
+                            merged_lower.add(m.lower())
+                    return merged
             # Use profile's fallback_models if defined
             if _p.fallback_models:
                 return list(_p.fallback_models)