fix(deepseek): drop retired models from picker and provider defaults

Stop offering deepseek-chat/reasoner in the static catalog and point
fallback/aux defaults at the permanent v4 IDs. Keep retired aliases in
a detection-only map so /model deepseek-chat still resolves to deepseek.

hermes_cli/models.py

@@ -390,8 +390,6 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
     "deepseek": [
         "deepseek-v4-pro",
         "deepseek-v4-flash",
-        "deepseek-chat",
-        "deepseek-reasoner",
     ],
     "xiaomi": [
         "mimo-v2.5-pro",
@@ -2085,11 +2083,25 @@ def _provider_keys(provider: str) -> set[str]:
     return {k for k in (key, normalized) if k}
 
 
+# Retired model IDs kept for /model auto-detect only — not shown in pickers.
+# DeepSeek cut these off on 2026-07-24; model_normalize remaps them on the wire.
+_PROVIDER_RETIRED_ALIASES: dict[str, tuple[str, ...]] = {
+    "deepseek": ("deepseek-chat", "deepseek-reasoner"),
+}
+
+
+def _provider_catalog_names(provider: str) -> tuple[str, ...]:
+    """Active picker models plus retired aliases recognized for detection."""
+    active = tuple(_PROVIDER_MODELS.get(provider, []))
+    retired = _PROVIDER_RETIRED_ALIASES.get(provider, ())
+    return active + retired
+
+
 def _model_in_provider_catalog(name_lower: str, providers: set[str]) -> bool:
     return any(
         name_lower == model.lower()
         for provider in providers
-        for model in _PROVIDER_MODELS.get(provider, [])
+        for model in _provider_catalog_names(provider)
     )
 
 
@@ -2235,7 +2247,7 @@ def detect_static_provider_for_model(
         current_provider == "custom"
         or current_provider.startswith("custom:")
     )
-    for pid, models in _PROVIDER_MODELS.items():
+    for pid in _PROVIDER_MODELS:
         if (
             pid in current_keys
             or pid in _AGGREGATOR_PROVIDERS
@@ -2244,7 +2256,7 @@ def detect_static_provider_for_model(
             continue
         if _is_custom_current:
             continue
-        if any(name_lower == m.lower() for m in models):
+        if any(name_lower == m.lower() for m in _provider_catalog_names(pid)):
             return (pid, name)
 
     # Borrow-list providers (re-expose other vendors' models) only after every
@@ -2252,7 +2264,7 @@ def detect_static_provider_for_model(
     for pid in _BORROWED_MODEL_PROVIDERS:
         if pid in current_keys:
             continue
-        if any(name_lower == m.lower() for m in _PROVIDER_MODELS.get(pid, [])):
+        if any(name_lower == m.lower() for m in _provider_catalog_names(pid)):
             return (pid, name)
 
     return None

plugins/model-providers/deepseek/__init__.py

@@ -1,11 +1,11 @@
 """DeepSeek provider profile.
 
-DeepSeek's V4 family (and the legacy ``deepseek-reasoner``) defaults to
-thinking-mode ON when ``extra_body.thinking`` is unset.  The API then returns
-``reasoning_content`` and starts enforcing the contract that subsequent turns
-echo it back; combined with how Hermes replays history this lands on the
-notorious HTTP 400 ``reasoning_content must be passed back`` error after the
-first tool call (#15700, #17212, #17825).
+DeepSeek's V4 family defaults to thinking-mode ON when ``extra_body.thinking``
+is unset.  The API then returns ``reasoning_content`` and starts enforcing
+the contract that subsequent turns echo it back; combined with how Hermes
+replays history this lands on the notorious HTTP 400
+``reasoning_content must be passed back`` error after the first tool call
+(#15700, #17212, #17825).
 
 This profile overrides :meth:`build_api_kwargs_extras` to mirror the Kimi /
 Moonshot wire shape that DeepSeek's OpenAI-compat endpoint expects:
@@ -13,8 +13,12 @@ Moonshot wire shape that DeepSeek's OpenAI-compat endpoint expects:
     {"reasoning_effort": "<low|medium|high|max>",
      "extra_body": {"thinking": {"type": "enabled" | "disabled"}}}
 
-Non-thinking models (only ``deepseek-chat`` today, which is V3) are left as
-no-ops so we don't perturb the V3 wire format.
+Non-thinking models (``deepseek-v3-*`` variants) are left as no-ops so we
+don't perturb the V3 wire format.
+
+The legacy aliases ``deepseek-chat`` / ``deepseek-reasoner`` were retired on
+2026-07-24.  Use ``deepseek-v4-flash`` or ``deepseek-v4-pro``; Hermes remaps
+the retired IDs in ``hermes_cli.model_normalize``.
 """
 
 from __future__ import annotations
@@ -29,8 +33,8 @@ def _model_supports_thinking(model: str | None) -> bool:
     """DeepSeek thinking-capable model families.
 
     Currently covers the V4 family (``deepseek-v4-pro``, ``deepseek-v4-flash``,
-    and any future ``deepseek-v4-*`` variants) and the legacy
-    ``deepseek-reasoner`` (R1).  ``deepseek-chat`` is V3 with no thinking mode.
+    and any future ``deepseek-v4-*`` variants).  Retired aliases are remapped
+    before requests leave Hermes, so they are not listed here.
     """
     m = (model or "").strip().lower()
     if not m:
@@ -39,8 +43,6 @@ def _model_supports_thinking(model: str | None) -> bool:
         # deepseek-v4-*, deepseek-v5-*, etc. — every V4+ generation has
         # thinking. v3 explicitly excluded.
         return True
-    if m == "deepseek-reasoner":
-        return True
     return False
 
 
@@ -90,11 +92,11 @@ deepseek = DeepSeekProfile(
     description="DeepSeek — native DeepSeek API",
     signup_url="https://platform.deepseek.com/",
     fallback_models=(
-        "deepseek-chat",
-        "deepseek-reasoner",
+        "deepseek-v4-pro",
+        "deepseek-v4-flash",
     ),
     base_url="https://api.deepseek.com/v1",
-    default_aux_model="deepseek-chat",
+    default_aux_model="deepseek-v4-flash",
 )
 
 register_provider(deepseek)

tests/hermes_cli/test_models.py

@@ -252,12 +252,18 @@ class TestDetectProviderForModel:
         assert result[0] == "anthropic"
 
     def test_deepseek_model_detected(self):
-        """deepseek-chat should resolve to deepseek provider."""
+        """Retired deepseek-chat alias still resolves to deepseek for /model."""
         result = detect_provider_for_model("deepseek-chat", "openai-codex")
         assert result is not None
         # Provider is deepseek (direct) or openrouter (fallback) depending on creds
         assert result[0] in {"deepseek", "openrouter"}
 
+    def test_deepseek_v4_model_detected(self):
+        """Current DeepSeek V4 IDs resolve to the deepseek provider."""
+        result = detect_provider_for_model("deepseek-v4-flash", "openai-codex")
+        assert result is not None
+        assert result[0] in {"deepseek", "openrouter"}
+
     def test_current_provider_model_returns_none(self):
         """Models belonging to the current provider should not trigger a switch."""
         assert detect_provider_for_model("gpt-5.3-codex", "openai-codex") is None

tests/plugins/model_providers/test_deepseek_profile.py

@@ -1,10 +1,10 @@
 """Unit tests for the DeepSeek provider profile's thinking-mode wiring.
 
-DeepSeek V4 (and the legacy ``deepseek-reasoner``) expects every request to
-carry an explicit ``extra_body.thinking`` parameter.  Omitting it makes the
-server default to thinking-mode ON, which then enforces the
-``reasoning_content``-must-be-echoed-back contract on subsequent turns and
-breaks the conversation with HTTP 400 (#15700, #17212, #17825).
+DeepSeek V4 expects every request to carry an explicit ``extra_body.thinking``
+parameter.  Omitting it makes the server default to thinking-mode ON, which
+then enforces the ``reasoning_content``-must-be-echoed-back contract on
+subsequent turns and breaks the conversation with HTTP 400 (#15700, #17212,
+#17825).
 
 These tests pin the profile's wire-shape contract so DeepSeek requests stay
 correctly shaped without going live.
@@ -106,7 +106,7 @@ class TestDeepSeekThinkingWireShape:
 
 
 class TestDeepSeekModelGating:
-    """V4 family + ``deepseek-reasoner`` get thinking; V3 stays untouched."""
+    """V4 family gets thinking; V3 / unknown stay untouched."""
 
     @pytest.mark.parametrize(
         "model",
@@ -114,7 +114,6 @@ class TestDeepSeekModelGating:
             "deepseek-v4-pro",
             "deepseek-v4-flash",
             "deepseek-v4-future-variant",
-            "deepseek-reasoner",
             "DEEPSEEK-V4-PRO",  # case-insensitive
         ],
     )
@@ -127,7 +126,6 @@ class TestDeepSeekModelGating:
     @pytest.mark.parametrize(
         "model",
         [
-            "deepseek-chat",         # V3 alias
             "deepseek-v3-0324",      # explicit V3
             "deepseek-v3.1",         # V3 minor revisions
             "",                       # bare/unknown
@@ -168,11 +166,11 @@ class TestDeepSeekFullKwargsIntegration:
         assert kwargs["reasoning_effort"] == "high"
         assert kwargs["extra_body"] == {"thinking": {"type": "enabled"}}
 
-    def test_v3_chat_full_kwargs_omit_thinking(self, deepseek_profile):
+    def test_v3_full_kwargs_omit_thinking(self, deepseek_profile):
         from agent.transports.chat_completions import ChatCompletionsTransport
 
         kwargs = ChatCompletionsTransport().build_kwargs(
-            model="deepseek-chat",
+            model="deepseek-v3-0324",
             messages=[{"role": "user", "content": "ping"}],
             tools=None,
             provider_profile=deepseek_profile,
@@ -195,12 +193,18 @@ class TestDeepSeekAuxModel:
     system.
     """
 
-    def test_profile_advertises_deepseek_chat(self, deepseek_profile):
-        assert deepseek_profile.default_aux_model == "deepseek-chat"
+    def test_profile_advertises_deepseek_v4_flash(self, deepseek_profile):
+        assert deepseek_profile.default_aux_model == "deepseek-v4-flash"
 
-    def test_consumer_api_returns_deepseek_chat(self):
+    def test_fallback_models_are_v4_only(self, deepseek_profile):
+        assert deepseek_profile.fallback_models == (
+            "deepseek-v4-pro",
+            "deepseek-v4-flash",
+        )
+
+    def test_consumer_api_returns_deepseek_v4_flash(self):
         from agent.auxiliary_client import _get_aux_model_for_provider
-        assert _get_aux_model_for_provider("deepseek") == "deepseek-chat"
+        assert _get_aux_model_for_provider("deepseek") == "deepseek-v4-flash"
 
     def test_consumer_api_returns_non_empty(self):
         from agent.auxiliary_client import _get_aux_model_for_provider