fix(deepseek): deepseek-flash is the canonical Flash id; retired names fold onto it

DeepSeek retired deepseek-v4-flash on 2026-09-10 (V4.1-Flash release); the API's
model name is now `deepseek-flash` and /v1/models lists only it. Hermes still
folded every non-V-series name onto deepseek-v4-flash, so `/model deepseek-flash`
on the DeepSeek provider was rewritten, then the validator "auto-corrected" it
back against the live listing: "Auto-corrected deepseek-v4-flash -> deepseek-flash"
on every switch.

Retired aliases (deepseek-chat / -reasoner and other fuzzy names) now fold onto
deepseek-flash; the curated catalog, profile fallback list, aux default, goal-judge
hint and pricing snapshot (2026-09-10 off-peak USD) follow the docs. Dated
deepseek-v4-* ids still pass through untouched.

Builds on YipTszkwan's #107126 (earliest fix in the cluster).

agent/usage_pricing.py

@@ -186,11 +186,11 @@ _SNAPSHOTS: tuple[tuple[str, Optional[str], str, dict], ...] = (
         "gpt-4.1-nano": ("0.10", "0.40", "0.025"), "o3": ("10.00", "40.00", "2.50"),
         "o3-mini": ("1.10", "4.40", "0.55"),
     }),
-    # deepseek-chat / deepseek-reasoner are deprecated aliases of
-    # deepseek-v4-flash's non-thinking / thinking modes — same rates.
-    ("deepseek", "https://api-docs.deepseek.com/quick_start/pricing", "deepseek-pricing-2026-07", {
-        ("deepseek-chat", "deepseek-reasoner", "deepseek-v4-flash"): ("0.14", "0.28", "0.0028"),
-        "deepseek-v4-pro": ("0.435", "0.87", "0.003625"),
+    # Off-peak USD rates (peak = 2x, Mon-Fri 01-04 + 06-10 UTC). ``deepseek-v4-flash`` and the
+    # retired deepseek-chat / deepseek-reasoner aliases are served by V4.1-Flash at the Flash price.
+    ("deepseek", "https://api-docs.deepseek.com/quick_start/pricing", "deepseek-pricing-2026-09-10", {
+        ("deepseek-flash", "deepseek-v4-flash", "deepseek-chat", "deepseek-reasoner"): ("0.15", "0.60", "0.003"),
+        "deepseek-v4-pro": ("0.66", "1.98", "0.022"),
     }),
     ("google", "https://ai.google.dev/gemini-api/docs/pricing", "google-pricing-2026-09-02", {
         ("gemini-3.8-flash", "gemini-3.7-flash"): ("0.75", "3.75", "0.075"),

hermes_cli/goals.py

@@ -1499,7 +1499,7 @@ class GoalManager:
                 f"judge API unreachable {n_tx} turns in a row (check auxiliary.goal_judge provider/key in config.yaml)",
                 "continue", reason,
                 f"⏸ Goal paused — judge API returned errors ({n_tx} turns). Check the goal_judge provider/key in "
-                + _JUDGE_CONFIG_HINT.format(provider="deepseek", model="deepseek-v4-flash"),
+                + _JUDGE_CONFIG_HINT.format(provider="deepseek", model="deepseek-flash"),
             )
         if n_parse >= DEFAULT_MAX_CONSECUTIVE_PARSE_FAILURES:
             return self._pause_decision(

hermes_cli/model_normalize.py

@@ -86,18 +86,17 @@ _CATALOGUE_PREFIX_REPAIR_PROVIDERS: frozenset[str] = frozenset({
 _LOWERCASE_MODEL_PROVIDERS: frozenset[str] = frozenset({
     "xiaomi"})
 
-# DeepSeek's direct API only accepts first-class V-series IDs after the 2026-07-24 cut-off (HTTP 400
-# otherwise). Both retired aliases map to deepseek-v4-flash per the official docs (thinking mode is
-# controlled by extra_body.thinking on the profile), so saved configs can't keep sending them.
+# DeepSeek's direct API only accepts first-class ids after the 2026-07-24 cut-off (HTTP 400
+# otherwise). Retired aliases fold onto the version-less ``deepseek-flash`` (V4.1-Flash, 2026-09-10;
+# thinking mode is controlled by extra_body.thinking on the profile), so saved configs can't keep
+# sending them.
 _DEEPSEEK_RETIRED_ALIASES: frozenset[str] = frozenset({
     "deepseek-chat", "deepseek-reasoner"})
 
-# ``deepseek-flash`` is the version-less canonical Flash id from the 2026-09 Flash refresh:
-# ``GET /v1/models`` reports it and the API accepts it directly. It carries no ``v<N>``
-# marker, so without an entry here it misses the V-series regex below and the id the user
-# picked is folded onto ``deepseek-v4-flash`` before it ever reaches the wire.
+# ``deepseek-flash`` carries no ``v<N>`` marker, so it needs an entry here or the V-series regex
+# below misses it and the id the user picked is rewritten before it reaches the wire.
 _DEEPSEEK_CANONICAL_MODELS: frozenset[str] = frozenset({
-    "deepseek-v4-pro", "deepseek-v4-flash", "deepseek-flash"})
+    "deepseek-flash", "deepseek-v4-pro"})
 
 # First-class V-series IDs incl. future ``deepseek-v5-*`` and dated variants
 # (``deepseek-v4-flash-20260423``): verified real model ids, NOT aliases of ``deepseek-chat``.
@@ -106,12 +105,12 @@ _DEEPSEEK_V_SERIES_RE = re.compile(r"^deepseek-v\d+([-.].+)?$")
 
 def _normalize_for_deepseek(model_name: str) -> str:
     """Map a model input to a DeepSeek-accepted id: canonicals and ``deepseek-v<digit>…`` pass
-    through (future V-series work without a release); retired aliases and everything else become
-    ``deepseek-v4-flash``."""
+    through (dated variants and future V-series work without a release); retired aliases and
+    everything else become ``deepseek-flash``."""
     bare = _strip_vendor_prefix(model_name).lower()
     if bare in _DEEPSEEK_CANONICAL_MODELS or _DEEPSEEK_V_SERIES_RE.match(bare):
         return bare
-    return "deepseek-v4-flash"
+    return "deepseek-flash"
 
 
 def _strip_vendor_prefix(model_name: str) -> str:

hermes_cli/models_catalog_static.py

@@ -205,7 +205,7 @@ _PROVIDER_MODELS: dict[str, list[str]] = {
         "claude-sonnet-4-6", "claude-opus-4-5-20251101", "claude-sonnet-4-5-20250929",
         "claude-opus-4-20250514", "claude-sonnet-4-20250514", "claude-haiku-4-5-20251001",
     ],
-    "deepseek": ["deepseek-v4-pro", "deepseek-v4-flash"],
+    "deepseek": ["deepseek-v4-pro", "deepseek-flash"],
     "xiaomi": ["mimo-v2.5-pro", "mimo-v2.5", "mimo-v2-pro", "mimo-v2-omni", "mimo-v2-flash"],
     "tencent-tokenhub": list(_TENCENT_MODELS),
     "tencent-tokenplan": list(_TENCENT_MODELS),

plugins/model-providers/deepseek/__init__.py

@@ -54,8 +54,8 @@ class DeepSeekProfile(ProviderProfile):
 deepseek = DeepSeekProfile(
     name="deepseek", aliases=("deepseek-chat",), env_vars=("DEEPSEEK_API_KEY",), display_name="DeepSeek",
     description="DeepSeek — native DeepSeek API", signup_url="https://platform.deepseek.com/",
-    fallback_models=("deepseek-v4-pro", "deepseek-v4-flash"), base_url="https://api.deepseek.com/v1",
-    default_aux_model="deepseek-v4-flash",
+    fallback_models=("deepseek-v4-pro", "deepseek-flash"), base_url="https://api.deepseek.com/v1",
+    default_aux_model="deepseek-flash",
 )
 
 register_provider(deepseek)

tests/agent/test_usage_pricing.py

@@ -123,11 +123,12 @@ def test_deepseek_v4_pro_pricing_entry_exists():
     )
 
     assert entry is not None
-    assert entry.input_cost_per_million is not None
-    assert entry.output_cost_per_million is not None
-    assert float(entry.input_cost_per_million) == 0.435
-    assert float(entry.output_cost_per_million) == 0.87
-    assert float(entry.cache_read_cost_per_million) == 0.003625
+    assert entry.source == "official_docs_snapshot"
+    # Pro is the premium tier: every rate must sit above the Flash row's.
+    flash = get_pricing_entry("deepseek-flash", provider="deepseek")
+    assert entry.input_cost_per_million > flash.input_cost_per_million
+    assert entry.output_cost_per_million > flash.output_cost_per_million
+    assert entry.cache_read_cost_per_million > flash.cache_read_cost_per_million
 
 
 def test_bundled_pricing_skips_endpoint_metadata(monkeypatch):
@@ -174,14 +175,13 @@ def test_unknown_model_falls_back_to_endpoint_metadata(monkeypatch):
 
 
 
-def test_deepseek_deprecated_aliases_price_as_v4_flash():
-    """Invariant: deepseek-chat / deepseek-reasoner are deprecated aliases for
-    deepseek-v4-flash's non-thinking / thinking modes (deprecation 2026-07-24)
-    — they must bill at identical rates to the flash entry, or sessions on the
-    legacy names over/under-report cost."""
-    flash = get_pricing_entry("deepseek-v4-flash", provider="deepseek")
+def test_deepseek_deprecated_aliases_price_as_flash():
+    """Invariant: deepseek-v4-flash / deepseek-chat / deepseek-reasoner are retired aliases
+    served by the current Flash model — they must bill at identical rates to the
+    ``deepseek-flash`` entry, or sessions on the legacy names over/under-report cost."""
+    flash = get_pricing_entry("deepseek-flash", provider="deepseek")
     assert flash is not None
-    for alias in ("deepseek-chat", "deepseek-reasoner"):
+    for alias in ("deepseek-v4-flash", "deepseek-chat", "deepseek-reasoner"):
         entry = get_pricing_entry(alias, provider="deepseek")
         assert entry is not None, alias
         assert entry.input_cost_per_million == flash.input_cost_per_million, alias

tests/hermes_cli/test_model_normalize.py

@@ -129,7 +129,7 @@ class TestDeepseekVSeriesPassThrough:
 # ── DeepSeek post-2026-07-24 alias remapping ───────────────────────────
 
 class TestDeepseekCanonicalAndReasonerMapping:
-    """Retired aliases and fuzzy names rewrite to deepseek-v4-flash.
+    """Retired aliases and fuzzy names rewrite to deepseek-flash.
 
     DeepSeek cut off ``deepseek-chat`` / ``deepseek-reasoner`` on
     2026-07-24; sending them on the wire returns HTTP 400.
@@ -139,7 +139,7 @@ class TestDeepseekCanonicalAndReasonerMapping:
     def test_provider_path_rewrites_reasoner(self):
         assert (
             normalize_model_for_provider("deepseek-reasoner", "deepseek")
-            == "deepseek-v4-flash"
+            == "deepseek-flash"
         )
 
     @pytest.mark.parametrize("model", [
@@ -149,8 +149,8 @@ class TestDeepseekCanonicalAndReasonerMapping:
         "deepseek-reasoning-preview",
         "deepseek-cot-experimental",
     ])
-    def test_reasoner_keywords_map_to_v4_flash(self, model):
-        assert _normalize_for_deepseek(model) == "deepseek-v4-flash"
+    def test_reasoner_keywords_map_to_flash(self, model):
+        assert _normalize_for_deepseek(model) == "deepseek-flash"
 
 
 # ── Regression: issue #78796 ───────────────────────────────────────────

tests/plugins/model_providers/test_deepseek_profile.py

@@ -190,16 +190,15 @@ class TestDeepSeekAuxModel:
     system.
     """
 
-    def test_profile_advertises_deepseek_v4_flash(self, deepseek_profile):
-        assert deepseek_profile.default_aux_model == "deepseek-v4-flash"
+    def test_profile_advertises_deepseek_flash(self, deepseek_profile):
+        assert deepseek_profile.default_aux_model == "deepseek-flash"
 
-    def test_fallback_models_are_v4_only(self, deepseek_profile):
-        assert deepseek_profile.fallback_models == (
-            "deepseek-v4-pro",
-            "deepseek-v4-flash",
-        )
+    def test_fallback_models_are_current_ids(self, deepseek_profile):
+        from hermes_cli.model_normalize import _normalize_for_deepseek
+        # Every advertised id must survive normalization unchanged (no retired alias in the picker).
+        assert all(_normalize_for_deepseek(m) == m for m in deepseek_profile.fallback_models)
 
-    def test_consumer_api_returns_deepseek_v4_flash(self):
+    def test_consumer_api_returns_deepseek_flash(self):
         from agent.auxiliary_client import _get_aux_model_for_provider
-        assert _get_aux_model_for_provider("deepseek") == "deepseek-v4-flash"
+        assert _get_aux_model_for_provider("deepseek") == "deepseek-flash"