feat: update model version from claude-sonnet-4-5 to claude-sonnet-4-6 and related adjustments

EvoScientist/cli/commands.py

@@ -102,7 +102,7 @@ def onboard(
     workspace settings, and agent parameters.
 
     Any answer can be pre-set via a flag (``--provider anthropic
-    --model claude-sonnet-4-5 ...``); prompts for unset answers stay
+    --model claude-sonnet-4-6 ...``); prompts for unset answers stay
     interactive unless ``--non-interactive`` is passed, in which case any
     missing required answer aborts the wizard.
     """

EvoScientist/config/onboard/wizard.py

@@ -572,7 +572,7 @@ def run_onboard(
                         # leaves a stale ``--model`` / ``--api-key`` that would
                         # be re-applied under a different provider, producing
                         # an invalid pair (e.g. ``provider=openai`` +
-                        # ``model=claude-sonnet-4-5``).
+                        # ``model=claude-sonnet-4-6``).
                         for field_name in vars(loop_snapshot):
                             setattr(
                                 config, field_name, getattr(loop_snapshot, field_name)

EvoScientist/config/settings.py

@@ -85,7 +85,7 @@ class EvoScientistConfig:
 
     # LLM Settings
     provider: str = "anthropic"
-    model: str = "claude-sonnet-4-5"
+    model: str = "claude-sonnet-4-6"
     model_fallbacks: str = ""  # "model:provider,model:provider" fallback chain
 
     # Async Sub-agent Settings

EvoScientist/llm/context_window.py

@@ -17,6 +17,11 @@ _KNOWN_MODEL_CONTEXT_WINDOWS: dict[str, int] = {
     # Qwen 3.6 open-source variants — exceptions to the ``qwen3.6`` family.
     "qwen3.6-27b": 262_000,
     "qwen3.6-35b-a3b": 262_000,
+    # Qwen 3.7 Max — closed-source flagship (1M).
+    "qwen3.7-max": 1_000_000,
+    # xAI Grok — per-model windows (build-0.1: 256K, 4.3: 1M).
+    "grok-build-0.1": 256_000,
+    "grok-4.3": 1_000_000,
     # Claude Haiku 4.5 — exception to the ``claude-`` family (200K, not 1M).
     "claude-haiku-4-5": 200_000,
 }
@@ -29,6 +34,8 @@ _KNOWN_MODEL_FAMILIES: list[tuple[str, int]] = [
     ("claude-", 1_000_000),
     # OpenAI GPT-5.5 family — base, pro, future variants
     ("gpt-5.5", 1_050_000),
+    # Google Gemini 3.x family — flash, flash-lite, pro (1.05M). Excludes 2.5.
+    ("gemini-3", 1_050_000),
     # Moonshot Kimi K2 family — k2.5, k2.6, k2-thinking, k2-thinking-turbo
     ("kimi-k2", 262_000),
     # Zhipu GLM-5 family — base, 5.1, 5-turbo, 5v-turbo, etc.

EvoScientist/llm/models.py

@@ -66,10 +66,9 @@ _THINKING_CAPABLE_PROVIDERS: set[str] = {"minimax"}
 # Model registry: list of (short_name, model_id, provider)
 # Allows same short_name across different providers.
 _MODEL_ENTRIES: list[tuple[str, str, str]] = [
-    # Custom Anthropic (third-party Claude-compatible endpoints, 3 defaults)
+    # Custom Anthropic (third-party Claude-compatible endpoints, current-gen defaults)
     # Listed BEFORE native anthropic so MODELS dict defaults to native provider
     ("claude-sonnet-4-6", "claude-sonnet-4-6", "custom-anthropic"),
-    ("claude-sonnet-4-5", "claude-sonnet-4-5", "custom-anthropic"),
     ("claude-haiku-4-5", "claude-haiku-4-5", "custom-anthropic"),
     # Custom OpenAI (third-party OpenAI-compatible endpoints, 3 defaults)
     # Listed BEFORE native openai so MODELS dict defaults to native provider
@@ -78,12 +77,9 @@ _MODEL_ENTRIES: list[tuple[str, str, str]] = [
     ("gpt-5.4", "gpt-5.4", "custom-openai"),
     ("gpt-5.3-codex", "gpt-5.3-codex", "custom-openai"),
     ("gpt-5-mini", "gpt-5-mini", "custom-openai"),
-    # Anthropic (ordered by capability)
+    # Anthropic (current generation)
-    ("claude-opus-4-7", "claude-opus-4-7", "anthropic"),
+    ("claude-opus-4-8", "claude-opus-4-8", "anthropic"),
-    ("claude-opus-4-6", "claude-opus-4-6", "anthropic"),
     ("claude-sonnet-4-6", "claude-sonnet-4-6", "anthropic"),
-    ("claude-opus-4-5", "claude-opus-4-5", "anthropic"),
-    ("claude-sonnet-4-5", "claude-sonnet-4-5", "anthropic"),
     ("claude-haiku-4-5", "claude-haiku-4-5", "anthropic"),
     # OpenAI
     ("gpt-5.5-pro", "gpt-5.5-pro", "openai"),
@@ -99,6 +95,7 @@ _MODEL_ENTRIES: list[tuple[str, str, str]] = [
     ("gpt-5-mini", "gpt-5-mini", "openai"),
     ("gpt-5-nano", "gpt-5-nano", "openai"),
     # Google GenAI
+    ("gemini-3.5-flash", "gemini-3.5-flash", "google-genai"),
     ("gemini-3.1-pro", "gemini-3.1-pro-preview", "google-genai"),
     (
         "gemini-3.1-pro-customtools",
@@ -133,13 +130,14 @@ _MODEL_ENTRIES: list[tuple[str, str, str]] = [
     ("kimi-k2.5", "Pro/moonshotai/Kimi-K2.5", "siliconflow"),
     ("glm-4.7", "Pro/zai-org/GLM-4.7", "siliconflow"),
     # OpenRouter
-    ("claude-opus-4.7", "anthropic/claude-opus-4.7", "openrouter"),
+    ("claude-opus-4.8", "anthropic/claude-opus-4.8", "openrouter"),
-    ("claude-opus-4.6", "anthropic/claude-opus-4.6", "openrouter"),
+    ("claude-opus-4.8-fast", "anthropic/claude-opus-4.8-fast", "openrouter"),
     ("claude-sonnet-4.6", "anthropic/claude-sonnet-4.6", "openrouter"),
     ("gpt-5.5-pro", "openai/gpt-5.5-pro", "openrouter"),
     ("gpt-5.5", "openai/gpt-5.5", "openrouter"),
     ("gpt-5.4", "openai/gpt-5.4", "openrouter"),
     ("gpt-5.3-codex", "openai/gpt-5.3-codex", "openrouter"),
+    ("gemini-3.5-flash", "google/gemini-3.5-flash", "openrouter"),
     ("gemini-3.1-pro", "google/gemini-3.1-pro-preview", "openrouter"),
     ("gemini-3-flash", "google/gemini-3-flash-preview", "openrouter"),
     ("kimi-k2.6", "moonshotai/kimi-k2.6", "openrouter"),
@@ -147,7 +145,10 @@ _MODEL_ENTRIES: list[tuple[str, str, str]] = [
     ("minimax-m2.7", "minimax/minimax-m2.7", "openrouter"),
     ("mimo-v2.5-pro", "xiaomi/mimo-v2.5-pro", "openrouter"),
     ("mimo-v2.5", "xiaomi/mimo-v2.5", "openrouter"),
-    ("grok-4.1-fast", "x-ai/grok-4.1-fast", "openrouter"),
+    ("grok-build-0.1", "x-ai/grok-build-0.1", "openrouter"),
+    ("grok-4.3", "x-ai/grok-4.3", "openrouter"),
+    ("qwen3.7-max", "qwen/qwen3.7-max", "openrouter"),
+    ("qwen3.6-flash", "qwen/qwen3.6-flash", "openrouter"),
     ("qwen3.5-122b", "qwen/qwen3.5-122b-a10b", "openrouter"),
     ("deepseek-v4-pro", "deepseek/deepseek-v4-pro", "openrouter"),
     ("deepseek-v4-flash", "deepseek/deepseek-v4-flash", "openrouter"),
@@ -246,7 +247,7 @@ def _apply_auto_config(
             _is_proxy = False
         if _is_proxy or (is_third_party and not _supports_thinking):
             pass
-        elif model_id.endswith(("4-6", "4-7")):
+        elif model_id.endswith(("4-6", "4-7", "4-8")):
             kwargs["thinking"] = {"type": "adaptive"}
             kwargs.setdefault("effort", "max")
         else:
@@ -282,8 +283,8 @@ def get_chat_model(
     """Get a chat model instance.
 
     Args:
-        model: Model name (short name like 'claude-sonnet-4-5' or full ID
+        model: Model name (short name like 'claude-sonnet-4-6' or full ID
-               like 'claude-sonnet-4-5-20250929'). Defaults to DEFAULT_MODEL.
+               like 'claude-sonnet-4-6-20250929'). Defaults to DEFAULT_MODEL.
         provider: Override the provider (e.g., 'anthropic', 'openai').
                   If not specified, inferred from model name or defaults to 'anthropic'.
         **kwargs: Additional arguments passed to init_chat_model (e.g., temperature).
@@ -292,8 +293,8 @@ def get_chat_model(
         A LangChain chat model instance.
 
     Examples:
-        >>> model = get_chat_model()  # Uses default (claude-sonnet-4-5)
+        >>> model = get_chat_model()  # Uses default (claude-sonnet-4-6)
-        >>> model = get_chat_model("claude-opus-4-5")  # Use short name
+        >>> model = get_chat_model("claude-opus-4-8")  # Use short name
         >>> model = get_chat_model("gpt-4o")  # OpenAI model
         >>> model = get_chat_model("claude-3-opus-20240229", provider="anthropic")  # Full ID
     """

tests/test_config.py

@@ -76,7 +76,7 @@ class TestEvoScientistConfig:
         assert config.openai_api_key == ""
         assert config.tavily_api_key == ""
         assert config.provider == "anthropic"
-        assert config.model == "claude-sonnet-4-5"
+        assert config.model == "claude-sonnet-4-6"
         assert config.default_mode == "daemon"
         assert config.default_workdir == ""
         assert config.show_thinking is True
@@ -157,7 +157,7 @@ class TestLoadSaveReset:
         """Test that load returns defaults when config file doesn't exist."""
         config = load_config()
         assert config.provider == "anthropic"
-        assert config.model == "claude-sonnet-4-5"
+        assert config.model == "claude-sonnet-4-6"
 
     def test_save_creates_file(self, temp_config_dir, clean_env):
         """Test that save creates the config file."""
@@ -350,8 +350,8 @@ class TestPriorityChain:
         """Test CLI arguments override file config."""
         save_config(EvoScientistConfig(model="gpt-4o"))
 
-        config = get_effective_config(cli_overrides={"model": "claude-opus-4-5"})
+        config = get_effective_config(cli_overrides={"model": "claude-opus-4-8"})
-        assert config.model == "claude-opus-4-5"
+        assert config.model == "claude-opus-4-8"
 
     def test_env_ui_backend_override(self, temp_config_dir, monkeypatch):
         """UI backend can be selected via environment variable."""

tests/test_configurable_model_middleware.py

@@ -196,7 +196,7 @@ class TestModelOverride:
 
         with (
             _patched_config(
-                {"model": "claude-opus-4-7", "model_provider": "anthropic"}
+                {"model": "claude-opus-4-8", "model_provider": "anthropic"}
             ),
             patch(
                 "EvoScientist.llm.get_chat_model", return_value=new_model
@@ -205,7 +205,7 @@ class TestModelOverride:
             result = _run(mw.awrap_model_call(req, handler))
 
         assert result == "ok"
-        mock_get.assert_called_once_with(model="claude-opus-4-7", provider="anthropic")
+        mock_get.assert_called_once_with(model="claude-opus-4-8", provider="anthropic")
 
     def test_provider_omitted_passed_as_none(self):
         mw = ConfigurableModelMiddleware()

tests/test_context_window.py

@@ -63,7 +63,7 @@ def test_resolve_context_window_falls_back_when_missing():
 
 
 def test_patch_table_resolves_known_model_name():
-    model = SimpleNamespace(model_name="claude-opus-4.7", profile=None)
+    model = SimpleNamespace(model_name="claude-opus-4.8", profile=None)
 
     assert get_context_window(model) == 1_000_000
 
@@ -78,7 +78,7 @@ def test_patch_table_strips_provider_prefix():
 
 def test_real_attribute_beats_patch_table():
     model = SimpleNamespace(
-        model_name="claude-opus-4.7",
+        model_name="claude-opus-4.8",
         context_window=500_000,
         profile=None,
     )
@@ -95,10 +95,10 @@ def test_unknown_model_falls_through_to_fallback():
 
 def test_claude_family_pattern_covers_all_variants():
     # Native Anthropic (dash form)
-    native = SimpleNamespace(model_name="claude-opus-4-7", profile=None)
+    native = SimpleNamespace(model_name="claude-opus-4-8", profile=None)
     # OpenRouter (dot form, with vendor prefix)
     openrouter = SimpleNamespace(model_name="anthropic/claude-sonnet-4.6", profile=None)
-    sonnet = SimpleNamespace(model_name="claude-sonnet-4-5", profile=None)
+    sonnet = SimpleNamespace(model_name="claude-sonnet-4-6", profile=None)
     # Haiku 4.5 is a dict-level exception (200K, not 1M).
     haiku = SimpleNamespace(model_name="claude-haiku-4-5", profile=None)
 
@@ -141,7 +141,7 @@ def test_claude_3x_matches_family_but_upstream_profile_wins_in_practice():
 
 
 def test_apply_injects_into_empty_profile():
-    model = SimpleNamespace(model_name="claude-opus-4-7", profile=None)
+    model = SimpleNamespace(model_name="claude-opus-4-8", profile=None)
 
     apply_known_context_window(model)
 
@@ -164,7 +164,7 @@ def test_apply_preserves_existing_profile_keys():
 
 def test_apply_does_not_overwrite_existing_max_input_tokens():
     model = SimpleNamespace(
-        model_name="claude-opus-4-7",
+        model_name="claude-opus-4-8",
         profile={"max_input_tokens": 500_000},
     )
 
@@ -185,7 +185,7 @@ def test_apply_handles_non_dict_profile_safely():
     # Non-dict profile (e.g. namespace from a future langchain version) must
     # not raise — fall back to a fresh dict so deepagents can still read it.
     model = SimpleNamespace(
-        model_name="claude-opus-4-7",
+        model_name="claude-opus-4-8",
         profile=SimpleNamespace(name="opaque object"),
     )
 

tests/test_llm.py

@@ -137,7 +137,7 @@ class TestListModels:
 class TestGetModelInfo:
     def test_returns_tuple_for_valid_model(self):
         """Test that get_model_info returns tuple for valid model."""
-        result = get_model_info("claude-sonnet-4-5")
+        result = get_model_info("claude-sonnet-4-6")
         assert result is not None
         assert isinstance(result, tuple)
         assert len(result) == 2
@@ -179,10 +179,10 @@ class TestGetChatModel:
         """Test that get_chat_model resolves short names correctly."""
         mock_init.return_value = "mock_model"
 
-        get_chat_model("claude-opus-4-5")
+        get_chat_model("claude-opus-4-8")
 
         call_kwargs = mock_init.call_args[1]
-        assert call_kwargs["model"] == "claude-opus-4-5"
+        assert call_kwargs["model"] == "claude-opus-4-8"
         assert call_kwargs["model_provider"] == "anthropic"
 
     @patch("EvoScientist.llm.models.init_chat_model")
@@ -213,7 +213,7 @@ class TestGetChatModel:
         """Test that provider can be overridden."""
         mock_init.return_value = "mock_model"
 
-        get_chat_model("claude-sonnet-4-5", provider="custom_provider")
+        get_chat_model("claude-sonnet-4-6", provider="custom_provider")
 
         call_kwargs = mock_init.call_args[1]
         assert call_kwargs["model_provider"] == "custom_provider"
@@ -403,7 +403,7 @@ class TestThirdPartyRouting:
         mock_init.return_value = "mock_model"
         monkeypatch.setenv("OPENROUTER_API_KEY", "or-key-456")
 
-        get_chat_model("x-ai/grok-4.1-fast", provider="openrouter")
+        get_chat_model("x-ai/grok-4.3", provider="openrouter")
 
         call_kwargs = mock_init.call_args[1]
         assert call_kwargs["model_provider"] == "openrouter"
@@ -417,7 +417,7 @@ class TestThirdPartyRouting:
         monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
 
         get_chat_model(
-            "x-ai/grok-4.1-fast",
+            "x-ai/grok-4.3",
             provider="openrouter",
             reasoning={"effort": "low"},
         )
@@ -432,7 +432,7 @@ class TestThirdPartyRouting:
         monkeypatch.setenv("OPENROUTER_API_KEY", "or-key")
         monkeypatch.setenv("EVOSCIENTIST_REASONING_EFFORT", "medium")
 
-        get_chat_model("x-ai/grok-4.1-fast", provider="openrouter")
+        get_chat_model("x-ai/grok-4.3", provider="openrouter")
 
         call_kwargs = mock_init.call_args[1]
         assert call_kwargs["reasoning"] == {"effort": "medium", "summary": "disabled"}
@@ -1176,7 +1176,7 @@ class TestAutoConfig:
         mock_init.return_value = "mock_model"
         monkeypatch.delenv("ANTHROPIC_BASE_URL", raising=False)
 
-        get_chat_model("claude-sonnet-4-5")
+        get_chat_model("claude-haiku-4-5")
 
         call_kwargs = mock_init.call_args[1]
         assert call_kwargs["thinking"] == {"type": "enabled", "budget_tokens": 10000}
@@ -1193,6 +1193,18 @@ class TestAutoConfig:
         assert call_kwargs["thinking"] == {"type": "adaptive"}
         assert call_kwargs["effort"] == "max"
 
+    @patch("EvoScientist.llm.models.init_chat_model")
+    def test_anthropic_4_8_adaptive_thinking(self, mock_init, monkeypatch):
+        """Anthropic 4-8 models get adaptive thinking with max effort."""
+        mock_init.return_value = "mock_model"
+        monkeypatch.delenv("ANTHROPIC_BASE_URL", raising=False)
+
+        get_chat_model("claude-opus-4-8")
+
+        call_kwargs = mock_init.call_args[1]
+        assert call_kwargs["thinking"] == {"type": "adaptive"}
+        assert call_kwargs["effort"] == "max"
+
     @patch("EvoScientist.llm.models.init_chat_model")
     def test_anthropic_4_6_proxy_no_thinking(self, mock_init, monkeypatch):
         """Anthropic 4-6 models via proxy skip thinking (history round-trip 422)."""
@@ -1213,7 +1225,7 @@ class TestAutoConfig:
         monkeypatch.setenv("ANTHROPIC_BASE_URL", "http://127.0.0.1:8000")
         monkeypatch.setenv("ANTHROPIC_API_KEY", "ccproxy-oauth")
 
-        get_chat_model("claude-sonnet-4-5")
+        get_chat_model("claude-haiku-4-5")
 
         call_kwargs = mock_init.call_args[1]
         assert "thinking" not in call_kwargs

tests/test_model_command.py

@@ -158,10 +158,10 @@ class TestModelCommandSwitch:
                 return_value=new_agent,
             ),
         ):
-            _run(cmd.execute(ctx, ["claude-opus-4-6"]))
+            _run(cmd.execute(ctx, ["claude-opus-4-8"]))
 
         # Config should be updated
-        assert cfg.model == "claude-opus-4-6"
+        assert cfg.model == "claude-opus-4-8"
         assert cfg.provider == "anthropic"
 
         # Agent should be replaced on context
@@ -170,7 +170,7 @@ class TestModelCommandSwitch:
         # Success message shown
         ui.append_system.assert_called_once()
         msg = ui.append_system.call_args[0][0]
-        assert "claude-opus-4-6" in msg
+        assert "claude-opus-4-8" in msg
         assert "anthropic" in msg
 
     def test_switch_with_save_flag(self):
@@ -198,10 +198,10 @@ class TestModelCommandSwitch:
             ),
             patch("EvoScientist.config.settings.set_config_value") as mock_save,
         ):
-            _run(cmd.execute(ctx, ["claude-opus-4-6", "--save"]))
+            _run(cmd.execute(ctx, ["claude-opus-4-8", "--save"]))
 
         # Config file should be updated
-        mock_save.assert_any_call("model", "claude-opus-4-6")
+        mock_save.assert_any_call("model", "claude-opus-4-8")
         mock_save.assert_any_call("provider", "anthropic")
 
         # Success message should mention save
@@ -233,7 +233,7 @@ class TestModelCommandSwitch:
             ),
             patch("EvoScientist.config.settings.set_config_value") as mock_save,
         ):
-            _run(cmd.execute(ctx, ["claude-opus-4-6"]))
+            _run(cmd.execute(ctx, ["claude-opus-4-8"]))
 
         # Config file should NOT be updated
         mock_save.assert_not_called()
@@ -273,7 +273,7 @@ class TestModelCommandFailure:
                 side_effect=RuntimeError("API key missing"),
             ) as mock_set,
         ):
-            _run(cmd.execute(ctx, ["claude-opus-4-6"]))
+            _run(cmd.execute(ctx, ["claude-opus-4-8"]))
 
         mock_set.assert_called_once()
         ui.append_system.assert_called_once()
@@ -558,7 +558,7 @@ class TestModelCommandLoadAgentFailure:
             # Pass ``--save`` to strengthen the assertion: if the ordering
             # ever regresses, ``set_config_value`` would be called with
             # stale data.
-            _run(cmd.execute(ctx, ["claude-opus-4-6", "--save"]))
+            _run(cmd.execute(ctx, ["claude-opus-4-8", "--save"]))
 
         # _load_agent was attempted (transactional first step).
         mock_load.assert_called_once()

tests/test_onboard.py

@@ -373,10 +373,10 @@ class TestStepModel:
         config = EvoScientistConfig()
 
         with patch("EvoScientist.config.onboard.steps.questionary") as mock_q:
-            mock_q.select.return_value.ask.return_value = "claude-sonnet-4-5"
+            mock_q.select.return_value.ask.return_value = "claude-sonnet-4-6"
             result = _step_model(config, "anthropic")
 
-        assert result == "claude-sonnet-4-5"
+        assert result == "claude-sonnet-4-6"
 
     def test_raises_keyboard_interrupt_on_cancel(self):
         """Test that _step_model raises KeyboardInterrupt on cancel."""
@@ -1142,7 +1142,7 @@ class TestRunOnboard:
                 "tui",  # UI backend
                 "anthropic",  # Provider
                 "api_key",  # Anthropic auth mode (API key, not OAuth)
-                "claude-sonnet-4-5",  # Model
+                "claude-sonnet-4-6",  # Model
                 "daemon",  # Workspace mode
                 True,  # Show thinking
             ]
@@ -1168,7 +1168,7 @@ class TestRunOnboard:
         # downstream prompts to consume the wrong values.
         final_config = mock_save.call_args_list[-1].args[0]
         assert final_config.provider == "anthropic"
-        assert final_config.model == "claude-sonnet-4-5"
+        assert final_config.model == "claude-sonnet-4-6"
         assert final_config.anthropic_auth_mode == "api_key"
         assert final_config.ui_backend == "tui"
         assert final_config.default_mode == "daemon"
@@ -1217,7 +1217,7 @@ class TestRunOnboard:
                 "tui",  # UI backend
                 "anthropic",  # Provider
                 "api_key",  # Anthropic auth mode
-                "claude-sonnet-4-5",  # Model
+                "claude-sonnet-4-6",  # Model
                 "daemon",  # Workspace mode
                 True,  # Show thinking
             ]
@@ -1287,7 +1287,7 @@ class TestRunOnboard:
                     "tui",
                     "anthropic",
                     "api_key",
-                    "claude-sonnet-4-5",
+                    "claude-sonnet-4-6",
                     "daemon",
                     True,
                 ]
@@ -1334,7 +1334,7 @@ class TestRunOnboard:
                 "tui",
                 "anthropic",
                 "api_key",
-                "claude-sonnet-4-5",
+                "claude-sonnet-4-6",
                 "daemon",
                 True,
             ]

tests/test_rich_command_ui.py

@@ -120,7 +120,7 @@ class TestUpdateStatusHook:
 
     def test_no_op(self):
         ui, console = _make_ui()
-        ui.update_status_after_model_change("claude-opus-4-6", "anthropic")
+        ui.update_status_after_model_change("claude-opus-4-8", "anthropic")
         console.print.assert_not_called()
 
 

tests/test_sessions.py

@@ -163,7 +163,7 @@ class TestThreadFunctions(unittest.TestCase):
                             "agent_name": AGENT_NAME,
                             "updated_at": f"2025-01-{15 + i}T10:00:00+00:00",
                             "workspace_dir": f"/tmp/ws_{tid}",
-                            "model": "claude-sonnet-4-5",
+                            "model": "claude-sonnet-4-6",
                         }
                     )
                     await conn.execute(
@@ -272,7 +272,7 @@ class TestThreadFunctions(unittest.TestCase):
         meta = _run(get_thread_metadata("abc12345"))
         assert meta is not None
         assert meta["workspace_dir"] == "/tmp/ws_abc12345"
-        assert meta["model"] == "claude-sonnet-4-5"
+        assert meta["model"] == "claude-sonnet-4-6"
 
     def test_get_thread_metadata_missing(self):
         meta = _run(get_thread_metadata("nonexist"))