feat(models): update the capable-tier Sonnet to Claude Sonnet 5

CHANGELOG.md

@@ -7,6 +7,15 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Changed
+
+- **Default "capable" model is now Claude Sonnet 5
+  (`claude-sonnet-5`).** Replaces `claude-sonnet-4-6` across the model
+  registry, adaptive routing, providers, telemetry, and token
+  estimation. Pricing is unchanged ($3/$15 per MTok standard;
+  introductory $2/$10 through 2026-08-31). 1M context, 128K max output,
+  adaptive thinking; `effort` defaults to `high`.
+
 ## [9.3.0] — 2026-06-30
 
 **Dynamic communication to improve the human/AI exchange.** Attune now

src/attune/agent_factory/base.py

@@ -307,11 +307,11 @@ def get_model_for_tier(self, tier: str, provider: str = "anthropic") -> str:
             defaults = {
                 "anthropic": {
                     "cheap": "claude-haiku-4-5",
-                    "capable": "claude-sonnet-4-6",
+                    "capable": "claude-sonnet-5",
                     "premium": "claude-opus-4-8",
                 },
             }
             return defaults.get(provider, defaults["anthropic"]).get(
                 tier,
-                "claude-sonnet-4-6",
+                "claude-sonnet-5",
             )

src/attune/agents/release/release_models.py

@@ -46,7 +46,7 @@
 
 MODEL_CONFIG = {
     "cheap": "claude-haiku-4-5",
-    "capable": "claude-sonnet-4-6",
+    "capable": "claude-sonnet-5",
     "premium": "claude-opus-4-8",
 }
 

src/attune/cli_commands/memory_agent.py

@@ -25,7 +25,7 @@
 
 # CAPABLE-tier stable alias — tracks the latest Sonnet minor, never retires
 # (avoids pinning a dated snapshot that EOLs).
-_DEFAULT_MODEL = "claude-sonnet-4-6"
+_DEFAULT_MODEL = "claude-sonnet-5"
 # Required beta header for the memory_20250818 tool + context management.
 _MEMORY_BETA = "context-management-2025-06-27"
 _DEFAULT_MAX_TOKENS = 4096

src/attune/cli_minimal.py

@@ -455,7 +455,7 @@ def _add_misc_subparsers(subparsers: argparse._SubParsersAction) -> None:
         "--path", help="Memory namespace root the agent addresses (default /memories)"
     )
     memory_agent_parser.add_argument(
-        "--model", help="Model id (default: claude-sonnet-4-6, the CAPABLE tier)"
+        "--model", help="Model id (default: claude-sonnet-5, the CAPABLE tier)"
     )
     memory_agent_parser.add_argument("--user-id", dest="user_id", help="Namespace memory per user")
     memory_agent_parser.add_argument(

src/attune/config/agent_config.py

@@ -149,14 +149,14 @@ def get_model_id(self) -> str:
         models = {
             Provider.ANTHROPIC: {
                 ModelTier.CHEAP: "claude-haiku-4-5",
-                ModelTier.CAPABLE: "claude-sonnet-4-6",
+                ModelTier.CAPABLE: "claude-sonnet-5",
                 ModelTier.PREMIUM: "claude-opus-4-8",
             },
         }
 
         return models.get(self.provider, {}).get(
             self.model_tier,
-            "claude-sonnet-4-6",  # Fallback
+            "claude-sonnet-5",  # Fallback
         )
 
     def for_book_production(self) -> "BookProductionConfig":

src/attune/config/sections/routing.py

@@ -33,7 +33,7 @@ class RoutingConfig:
 
     default_tier: Literal["cheap", "capable", "premium"] = "capable"
     cheap_model: str = "claude-haiku-4-5"
-    capable_model: str = "claude-sonnet-4-6"
+    capable_model: str = "claude-sonnet-5"
     premium_model: str = "claude-opus-4-8"
     auto_tier_selection: bool = True
     cost_optimization: bool = True
@@ -67,7 +67,7 @@ def from_dict(cls, data: dict) -> "RoutingConfig":
         return cls(
             default_tier=data.get("default_tier", "capable"),
             cheap_model=data.get("cheap_model", "claude-haiku-4-5"),
-            capable_model=data.get("capable_model", "claude-sonnet-4-6"),
+            capable_model=data.get("capable_model", "claude-sonnet-5"),
             premium_model=data.get("premium_model", "claude-opus-4-8"),
             auto_tier_selection=data.get("auto_tier_selection", True),
             cost_optimization=data.get("cost_optimization", True),

src/attune/config/xml_config.py

@@ -56,8 +56,8 @@ class AdaptiveConfig:
     model_tier_mapping: dict[str, str] = field(
         default_factory=lambda: {
             "simple": "claude-haiku-4-5",
-            "moderate": "claude-sonnet-4-6",
-            "complex": "claude-sonnet-4-6",
+            "moderate": "claude-sonnet-5",
+            "complex": "claude-sonnet-5",
             "very_complex": "claude-opus-4-8",
         },
     )

src/attune/help/polish.py

@@ -113,7 +113,7 @@ def _call_llm(
         # Stable alias — claude-sonnet-4-20250514 retires 2026-06-15.
         # The stable alias routes to the same checkpoint and stays
         # valid across model upgrades.
-        model="claude-sonnet-4-6",
+        model="claude-sonnet-5",
         max_tokens=4096,
         system=_SYSTEM_PROMPT,
         messages=[{"role": "user", "content": user_message}],

src/attune/llm/core.py

@@ -231,7 +231,7 @@ def _create_provider(
         if provider == "anthropic":
             return AnthropicProvider(
                 api_key=api_key,
-                model=model or "claude-sonnet-4-6",
+                model=model or "claude-sonnet-5",
                 **kwargs,
             )
         raise ValueError(f"Unknown provider: {provider}")

src/attune/llm/providers/anthropic.py

@@ -74,7 +74,7 @@ class AnthropicProvider(BaseLLMProvider):
     def __init__(
         self,
         api_key: str | None = None,
-        model: str = "claude-sonnet-4-6",
+        model: str = "claude-sonnet-5",
         use_prompt_caching: bool = False,  # Deprecated
         use_thinking: bool = False,  # Deprecated
         thinking_budget: int = 10000,
@@ -411,7 +411,7 @@ def get_model_info(self) -> dict[str, Any]:
                 "supports_thinking": True,
                 "ideal_for": "Complex reasoning, large codebases",
             },
-            "claude-sonnet-4-6": {
+            "claude-sonnet-5": {
                 "max_tokens": 200000,
                 "cost_per_1m_input": 3.00,
                 "cost_per_1m_output": 15.00,

src/attune/llm/providers/anthropic_batch.py

@@ -77,7 +77,7 @@ def create_batch(self, requests: list[dict[str, Any]], job_id: str | None = None
             ...     {
             ...         "custom_id": "task_1",
             ...         "params": {
-            ...             "model": "claude-sonnet-4-6",
+            ...             "model": "claude-sonnet-5",
             ...             "messages": [{"role": "user", "content": "Test"}],
             ...             "max_tokens": 1024
             ...         }
@@ -99,7 +99,7 @@ def create_batch(self, requests: list[dict[str, Any]], job_id: str | None = None
                 formatted_req = {
                     "custom_id": req.get("custom_id", f"req_{id(req)}"),
                     "params": {
-                        "model": req.get("model", "claude-sonnet-4-6"),
+                        "model": req.get("model", "claude-sonnet-5"),
                         "messages": req.get("messages", []),
                         "max_tokens": req.get("max_tokens", 4096),
                     },

src/attune/metrics/prompt_metrics.py

@@ -31,7 +31,7 @@ class PromptMetrics:
         workflow: Workflow name (e.g., "code_review", "bug_predict")
         agent_role: Role of the agent (e.g., "Code Reviewer")
         task_description: Brief description of task (truncated to 100 chars)
-        model: Model name (e.g., "claude-sonnet-4-6", "claude-haiku-4-5")
+        model: Model name (e.g., "claude-sonnet-5", "claude-haiku-4-5")
         prompt_tokens: Number of input tokens
         completion_tokens: Number of output tokens
         total_tokens: Total tokens (input + output)

src/attune/models/adaptive_routing.py

@@ -146,7 +146,7 @@ def _get_default_model(self, tier: str = "CHEAP") -> str:
         # Fallback to known models if registry lookup fails
         fallbacks = {
             "cheap": "claude-haiku-4-5",
-            "capable": "claude-sonnet-4-6",
+            "capable": "claude-sonnet-5",
             "premium": "claude-opus-4-8",
         }
         return fallbacks.get(tier_lower, "claude-haiku-4-5")

src/attune/models/executor.py

@@ -27,7 +27,7 @@ class LLMResponse:
 
     Attributes:
         content: The LLM response text
-        model_id: Model identifier (e.g., "claude-sonnet-4-6")
+        model_id: Model identifier (e.g., "claude-sonnet-5")
         provider: Provider name (e.g., "anthropic")
         tier: Model tier ("cheap", "capable", "premium")
         tokens_input: Number of input tokens used

src/attune/models/registry.py

@@ -125,7 +125,7 @@ def to_cost_tracker_pricing(self) -> dict[str, float]:
 MODEL_REGISTRY: dict[str, dict[str, ModelInfo]] = {
     # -------------------------------------------------------------------------
     # Anthropic Claude Models
-    # Intelligent fallback: Sonnet 4.6 → Opus 4.8 (~1.67x input cost: $3 → $5/M)
+    # Intelligent fallback: Sonnet 5 → Opus 4.8 (~1.67x input cost: $3 → $5/M)
     # -------------------------------------------------------------------------
     "anthropic": {
         "cheap": ModelInfo(
@@ -139,7 +139,7 @@ def to_cost_tracker_pricing(self) -> dict[str, float]:
             supports_tools=True,
         ),
         "capable": ModelInfo(
-            id="claude-sonnet-4-6",
+            id="claude-sonnet-5",
             provider="anthropic",
             tier="capable",
             input_cost_per_million=3.00,
@@ -177,7 +177,7 @@ class ModelRegistry:
         >>> registry = ModelRegistry()
         >>> model = registry.get_model("anthropic", "capable")
         >>> print(model.id)
-        claude-sonnet-4-6
+        claude-sonnet-5
 
         >>> models = registry.get_models_by_tier("cheap")
         >>> print(len(models))
@@ -239,7 +239,7 @@ def get_model(self, provider: str, tier: str) -> ModelInfo | None:
             >>> registry = ModelRegistry()
             >>> model = registry.get_model("anthropic", "capable")
             >>> print(model.id)
-            claude-sonnet-4-6
+            claude-sonnet-5
 
         """
         if provider.lower() != "anthropic":
@@ -360,7 +360,7 @@ def get_pricing_for_model(self, model_id: str) -> dict[str, float] | None:
 
         Example:
             >>> registry = ModelRegistry()
-            >>> pricing = registry.get_pricing_for_model("claude-sonnet-4-6")
+            >>> pricing = registry.get_pricing_for_model("claude-sonnet-5")
             >>> print(pricing)
             {'input': 3.0, 'output': 15.0}
 
@@ -463,6 +463,9 @@ def get_tiers() -> list[str]:
 
 TIER_PRICING: dict[str, dict[str, float]] = {
     "cheap": {"input": 1.00, "output": 5.00},  # Haiku 4.5 pricing
-    "capable": {"input": 3.00, "output": 15.00},  # Sonnet 4.6 pricing
+    "capable": {
+        "input": 3.00,
+        "output": 15.00,
+    },  # Sonnet 5 pricing (std; intro $2/$10 until 2026-08-31)
     "premium": {"input": 5.00, "output": 25.00},  # Opus 4.8 pricing
 }

src/attune/models/telemetry/analytics.py

@@ -202,7 +202,7 @@ def sonnet_opus_fallback_analysis(
             c
             for c in calls
             if c.provider == "anthropic"
-            and c.model_id in ["claude-sonnet-4-6", "claude-opus-4-6", "claude-opus-4-8"]
+            and c.model_id in ["claude-sonnet-5", "claude-opus-4-6", "claude-opus-4-8"]
         ]
 
         if not anthropic_calls:
@@ -222,7 +222,7 @@ def sonnet_opus_fallback_analysis(
         total = len(anthropic_calls)
 
         # Count Sonnet attempts and successes
-        sonnet_calls = [c for c in anthropic_calls if c.model_id == "claude-sonnet-4-6"]
+        sonnet_calls = [c for c in anthropic_calls if c.model_id == "claude-sonnet-5"]
         sonnet_successes = sum(1 for c in sonnet_calls if c.success)
 
         # Count Opus fallbacks (calls with fallback_used and ended up on Opus)

src/attune/models/token_estimator.py

@@ -43,7 +43,7 @@ def _get_encoding(model_id: str) -> Any:
     return tiktoken.get_encoding("cl100k_base")
 
 
-def estimate_tokens(text: str, model_id: str = "claude-sonnet-4-6") -> int:
+def estimate_tokens(text: str, model_id: str = "claude-sonnet-5") -> int:
     """Estimate token count for text using accurate token counting.
 
     Uses empathy_llm_toolkit's token counter which leverages tiktoken for fast,

src/attune/orchestration/tools/test_generation.py

@@ -210,7 +210,7 @@ class Foo:
             # latest Sonnet checkpoint. The old claude-3-5-sonnet-2024xxxx
             # dated snapshots were retired by Anthropic 2025-10-28.
             models_to_try = [
-                "claude-sonnet-4-6",  # Sonnet 4.6 (stable alias)
+                "claude-sonnet-5",  # Sonnet 5 (stable alias)
             ]
 
             response = None

src/attune/orchestration/tools/testing.py

@@ -356,7 +356,7 @@ class Foo:
             # latest Sonnet checkpoint. The old claude-3-5-sonnet-2024xxxx
             # dated snapshots were retired by Anthropic 2025-10-28.
             models_to_try = [
-                "claude-sonnet-4-6",  # Sonnet 4.6 (stable alias)
+                "claude-sonnet-5",  # Sonnet 5 (stable alias)
             ]
 
             response = None

src/attune/routing/model_router.py

@@ -180,7 +180,7 @@ def route(
             >>> router.route("summarize")
             'claude-haiku-4-5'
             >>> router.route("fix_bug")
-            'claude-sonnet-4-6'
+            'claude-sonnet-5'
             >>> router.route("coordinate")
             'claude-opus-4-8'
 

src/attune/template_defs_basic.py

@@ -26,7 +26,7 @@
   provider: "anthropic"
   models:
     cheap: "claude-haiku-4-5"
-    capable: "claude-sonnet-4-6"
+    capable: "claude-sonnet-5"
     premium: "claude-opus-4-8"
 
 # Claude Code integration
@@ -82,7 +82,7 @@
   provider: "anthropic"
   models:
     cheap: "claude-haiku-4-5"
-    capable: "claude-sonnet-4-6"
+    capable: "claude-sonnet-5"
     premium: "claude-opus-4-8"
 
 claude_sync:

src/attune/template_defs_web.py

@@ -25,7 +25,7 @@
   provider: "anthropic"
   models:
     cheap: "claude-haiku-4-5"
-    capable: "claude-sonnet-4-6"
+    capable: "claude-sonnet-5"
     premium: "claude-opus-4-8"
 
 claude_sync:
@@ -199,7 +199,7 @@ async def health():
   provider: "anthropic"
   models:
     cheap: "claude-haiku-4-5"
-    capable: "claude-sonnet-4-6"
+    capable: "claude-sonnet-5"
     premium: "claude-opus-4-8"
   task_overrides:
     summarize: "cheap"

src/attune/utils/tokens.py

@@ -94,7 +94,7 @@ def _count_tokens_heuristic(text: str) -> int:
     return max(1, len(text) // 4)
 
 
-def count_tokens(text: str, model: str = "claude-sonnet-4-6", use_api: bool = False) -> int:
+def count_tokens(text: str, model: str = "claude-sonnet-5", use_api: bool = False) -> int:
     """Count tokens using best available method.
 
     By default, uses tiktoken for fast local estimation (~98% accurate).
@@ -149,7 +149,7 @@ def count_tokens(text: str, model: str = "claude-sonnet-4-6", use_api: bool = Fa
 def count_message_tokens(
     messages: list[dict[str, str]],
     system_prompt: str | None = None,
-    model: str = "claude-sonnet-4-6",
+    model: str = "claude-sonnet-5",
     use_api: bool = False,
 ) -> dict[str, int]:
     """Count tokens in a conversation.
@@ -231,7 +231,7 @@ def count_message_tokens(
     return counts
 
 
-def estimate_cost(input_tokens: int, output_tokens: int, model: str = "claude-sonnet-4-6") -> float:
+def estimate_cost(input_tokens: int, output_tokens: int, model: str = "claude-sonnet-5") -> float:
     """Estimate cost in USD based on token counts.
 
     Args:
@@ -275,7 +275,7 @@ def calculate_cost_with_cache(
     output_tokens: int,
     cache_creation_tokens: int,
     cache_read_tokens: int,
-    model: str = "claude-sonnet-4-6",
+    model: str = "claude-sonnet-5",
 ) -> dict[str, Any]:
     """Calculate cost including Anthropic prompt caching.
 

src/attune/workflows/config.py

@@ -560,7 +560,7 @@ def create_example_config() -> str:
 custom_models:
   anthropic:
     cheap: claude-haiku-4-5
-    capable: claude-sonnet-4-6
+    capable: claude-sonnet-5
     premium: claude-opus-4-8
   openai:
     cheap: gpt-4o-mini
@@ -573,7 +573,7 @@ def create_example_config() -> str:
   # Create your own hybrid mix:
   hybrid:
     cheap: gpt-4o-mini           # OpenAI - cheapest per token
-    capable: claude-sonnet-4-6   # Anthropic - best code/reasoning
+    capable: claude-sonnet-5   # Anthropic - best code/reasoning
     premium: claude-opus-4-8   # Anthropic - best overall
 
 # =============================================================================
@@ -654,6 +654,6 @@ def create_example_config() -> str:
 # EMPATHY_WORKFLOW_PROVIDER=hybrid             # Default provider
 # EMPATHY_WORKFLOW_RESEARCH_PROVIDER=anthropic # Per-workflow
 # EMPATHY_MODEL_CHEAP=gpt-4o-mini              # Tier model override
-# EMPATHY_MODEL_CAPABLE=claude-sonnet-4-6
+# EMPATHY_MODEL_CAPABLE=claude-sonnet-5
 # EMPATHY_MODEL_PREMIUM=claude-opus-4-8
 """