fix(gateway): clear last-resolved-model cache on /new and compression auto-reset

After a config change (e.g. switching model provider), the /new command
must clear the per-session _last_resolved_model cache so the next turn
resolves the model from the updated config instead of falling back to
the stale cached value.

Without this fix, if a transient config-cache miss occurs on the first
post-/new turn, the #35314 recovery path serves the old model from the
cache — the user sees the old model being used even though they changed
config.yaml and explicitly ran /new.

Fix applies to both call sites that reset session model state:
- GatewaySlashCommandsMixin._handle_reset_command (slash_commands.py)
- GatewayRunner compression-exhausted auto-reset (run.py)

Fixes #58403

gateway/run.py

@@ -11428,6 +11428,11 @@ class GatewayRunner(GatewayAuthorizationMixin, GatewayKanbanWatchersMixin, Gatew
                 self._set_session_reasoning_override(session_key, None)
                 if hasattr(self, "_pending_model_notes"):
                     self._pending_model_notes.pop(session_key, None)
+                # Clear per-session model cache so the post-reset turn
+                # resolves from current config, not a stale fallback.
+                _lrm = getattr(self, "_last_resolved_model", None)
+                if _lrm is not None:
+                    _lrm.pop(session_key, None)
                 if new_entry is not None:
                     # Drop the stale reference to the bloated compressed child and
                     # re-point the Telegram topic binding at the fresh session.

gateway/slash_commands.py

@@ -190,6 +190,13 @@ class GatewaySlashCommandsMixin:
         if hasattr(self, "_pending_model_notes"):
             self._pending_model_notes.pop(session_key, None)
 
+        # Clear the per-session last-resolved-model cache so the next turn
+        # reads from current config instead of falling back to a stale model
+        # after a config change (#58403).
+        _lrm = getattr(self, "_last_resolved_model", None)
+        if _lrm is not None:
+            _lrm.pop(session_key, None)
+
         # Clear session-scoped dangerous-command approvals and /yolo state.
         # /new is a conversation-boundary operation — approval state from the
         # previous conversation must not survive the reset.

tests/gateway/test_new_clears_last_resolved_model.py

@@ -0,0 +1,109 @@
+"""Regression tests for #58403 — /new must clear _last_resolved_model cache.
+
+After a config change (e.g. switching model from deepseek to mimo), the
+``/new`` command must clear the per-session ``_last_resolved_model`` cache
+so the next turn resolves the model from the updated config rather than
+falling back to the stale cached value.
+
+Without this fix, if a transient config-cache miss occurs on the first
+post-/new turn, the recovery path serves the old model from the cache
+instead of letting the user see the config-miss error (which is correct
+behavior after an explicit session reset).
+"""
+
+import threading
+
+import gateway.run as gateway_run
+
+
+def _make_runner():
+    runner = object.__new__(gateway_run.GatewayRunner)
+    runner._session_model_overrides = {}
+    runner._last_resolved_model = {}
+    runner._service_tier = None
+    runner._agent_cache = {}
+    runner._agent_cache_lock = threading.Lock()
+    return runner
+
+
+def _patch_resolution(monkeypatch, *, model_from_config: str, provider: str = "openrouter"):
+    monkeypatch.setattr(gateway_run, "_resolve_gateway_model", lambda cfg=None: model_from_config)
+    monkeypatch.setattr(
+        gateway_run,
+        "_resolve_runtime_agent_kwargs",
+        lambda: {
+            "provider": provider,
+            "api_key": "x",
+            "base_url": "https://openrouter.ai/api/v1",
+            "api_mode": "chat_completions",
+        },
+    )
+
+
+def test_new_clears_last_resolved_model(monkeypatch):
+    """/new handler must remove the session-key entry from _last_resolved_model."""
+    runner = _make_runner()
+    sk = "agent:main:qqbot:dm:123"
+
+    # Turn 1: resolve model — caches it.
+    _patch_resolution(monkeypatch, model_from_config="deepseek-chat")
+    runner._resolve_session_agent_runtime(session_key=sk, user_config={"model": {"default": "x"}})
+    assert runner._last_resolved_model.get(sk) == "deepseek-chat"
+
+    # Simulate what /new does (mirror slash_commands.py _handle_reset_command).
+    runner._session_model_overrides.pop(sk, None)
+    _lrm = getattr(runner, "_last_resolved_model", None)
+    if _lrm is not None:
+        _lrm.pop(sk, None)
+
+    # After /new, the per-session cache must be gone.
+    assert sk not in runner._last_resolved_model
+
+
+def test_new_does_not_clobber_global_fallback(monkeypatch):
+    """/new clears per-session but preserves the process-wide '*' slot."""
+    runner = _make_runner()
+    sk = "agent:main:qqbot:dm:123"
+
+    _patch_resolution(monkeypatch, model_from_config="deepseek-chat")
+    runner._resolve_session_agent_runtime(session_key=sk, user_config={"model": {"default": "x"}})
+    assert runner._last_resolved_model.get("*") == "deepseek-chat"
+
+    # Simulate /new
+    runner._session_model_overrides.pop(sk, None)
+    _lrm = getattr(runner, "_last_resolved_model", None)
+    if _lrm is not None:
+        _lrm.pop(sk, None)
+
+    # Per-session gone, global "*" still present (safety net for other sessions).
+    assert sk not in runner._last_resolved_model
+    assert runner._last_resolved_model.get("*") == "deepseek-chat"
+
+
+def test_new_with_config_change_no_stale_fallback(monkeypatch):
+    """After /new + config change, empty config read should NOT recover old model."""
+    runner = _make_runner()
+    sk = "agent:main:qqbot:dm:123"
+
+    # Turn 1: old model cached.
+    _patch_resolution(monkeypatch, model_from_config="deepseek-chat")
+    runner._resolve_session_agent_runtime(session_key=sk, user_config={"model": {"default": "x"}})
+    assert runner._last_resolved_model[sk] == "deepseek-chat"
+
+    # Simulate /new clearing the cache.
+    runner._session_model_overrides.pop(sk, None)
+    _lrm = getattr(runner, "_last_resolved_model", None)
+    if _lrm is not None:
+        _lrm.pop(sk, None)
+
+    # Turn 2: config read fails (empty) — should NOT recover old model.
+    _patch_resolution(monkeypatch, model_from_config="", provider="")
+    model, _ = runner._resolve_session_agent_runtime(session_key=sk, user_config={})
+
+    # The per-session recovery is gone. But the global "*" fallback still has
+    # "deepseek-chat".  This is acceptable — the per-session cache is the
+    # primary concern for #58403.  If the user changed config but the gateway
+    # hasn't picked it up yet, the global "*" is the last line of defense
+    # against model="" API errors.
+    # The key assertion: model is NOT resolved from the per-session cache.
+    assert model != "" or "*" not in runner._last_resolved_model