models: add Claude Fable 5 and Opus 5, retire dead Sonnet 4 snapshot

CLAUDE.md

@@ -115,4 +115,4 @@ Docker image → GHCR (`ghcr.io/c4g/va-dat`) → Coolify at `https://va-dat.c4g.
 - **Silent dry runs.** A request with no resolvable key returns `200 OK` with no LLM findings and no CSV. The only signal is `summary.dry_run` in the response.
 - HTML test files can be enormous (500K+ tokens) — do not read them fully into context.
 - Some files may have non-UTF-8 encoding; always use `errors='replace'`.
-- Model defaults differ by entry point: `run_pipeline.py` uses `claude-sonnet-4-20250514`, `api_server.py` uses `claude-haiku-4-5-20251001`.
+- Model defaults differ by entry point: `run_pipeline.py` uses `claude-sonnet-5`, `api_server.py` uses `claude-haiku-4-5-20251001`.

README.md

@@ -313,7 +313,7 @@ uv run python entry_points/generate_report.py --output-dir ./output --report-dir
 |------|---------|-------------|
 | `--html` | (required) | Path to the HTML file to analyze |
 | `--output-dir` | `./output` | Directory for results |
-| `--model` | `claude-sonnet-4-20250514` | Anthropic model to use |
+| `--model` | `claude-sonnet-5` | Anthropic model to use |
 | `--dry-run` | off | Generate prompts without calling the API |
 | `--include-summaries` | off | Include the 3 cross-cutting summary prompts |
 | `--show-cost` | off | Print estimated dollar cost of the run based on model pricing |

entry_points/generate_report.py

@@ -880,13 +880,17 @@ def deduplicate_with_llm(
             response_text = response.choices[0].message.content or "[]"
         else:
             import anthropic
+            from processing_scripts.llm_client.client import supports_temperature
+
             client = anthropic.Anthropic(api_key=api_key)
-            message = client.messages.create(
-                model=model,
-                max_tokens=512,
-                temperature=0.0,
-                messages=[{"role": "user", "content": prompt}],
-            )
+            request_kwargs = {
+                "model": model,
+                "max_tokens": 512,
+                "messages": [{"role": "user", "content": prompt}],
+            }
+            if supports_temperature(model):
+                request_kwargs["temperature"] = 0.0
+            message = client.messages.create(**request_kwargs)
             response_text = "".join(
                 block.text for block in message.content if block.type == "text"
             )
@@ -1069,8 +1073,8 @@ def main():
     )
     parser.add_argument(
         "--model",
-        default="claude-sonnet-4-20250514",
-        help="Model to use for LLM deduplication (default: claude-sonnet-4-20250514)",
+        default="claude-sonnet-5",
+        help="Model to use for LLM deduplication (default: claude-sonnet-5)",
     )
     args = parser.parse_args()
 

entry_points/run_pipeline.py

@@ -53,6 +53,9 @@
 
 # Pricing per million tokens: (input, output)
 MODEL_PRICING = {
+    "claude-fable-5": (10.00, 50.00),
+    "claude-opus-5": (5.00, 25.00),
+    "claude-sonnet-5": (3.00, 15.00),
     "claude-sonnet-4": (3.00, 15.00),
     "claude-opus-4": (15.00, 75.00),
     "claude-haiku-4": (0.80, 4.00),
@@ -149,12 +152,17 @@ def call(self, prompt: str) -> dict:
 
     def _call_anthropic(self, prompt: str, start: float) -> dict:
         """Call the Anthropic Messages API."""
-        message = self._client.messages.create(
-            model=self.model,
-            max_tokens=self.max_tokens,
-            temperature=0.1,
-            messages=[{"role": "user", "content": prompt}],
-        )
+        from processing_scripts.llm_client.client import supports_temperature
+
+        request_kwargs = {
+            "model": self.model,
+            "max_tokens": self.max_tokens,
+            "messages": [{"role": "user", "content": prompt}],
+        }
+        if supports_temperature(self.model):
+            request_kwargs["temperature"] = 0.1
+
+        message = self._client.messages.create(**request_kwargs)
 
         response_text = "".join(
             block.text for block in message.content if block.type == "text"
@@ -519,8 +527,11 @@ def main():
     parser.add_argument(
         "--model",
         type=str,
-        default="claude-sonnet-4-20250514",
-        help="Model to use (default: claude-sonnet-4-20250514)",
+        default="claude-sonnet-5",
+        help=(
+            "Model to use (default: claude-sonnet-5). Accepts any "
+            "Claude, OpenAI, or Gemini model ID, e.g. claude-fable-5."
+        ),
     )
     parser.add_argument(
         "--dry-run",

index.html

@@ -550,8 +550,14 @@ <h2 id="audit-heading">Accessibility Audit Tool</h2>
                     <option value="claude-haiku-4-5-20251001">
                       Claude Haiku 4.5 — faster, cheaper
                     </option>
-                    <option value="claude-sonnet-4-20250514">
-                      Claude Sonnet 4 — higher quality
+                    <option value="claude-sonnet-5">
+                      Claude Sonnet 5 — higher quality
+                    </option>
+                    <option value="claude-opus-5">
+                      Claude Opus 5 — complex agentic/coding tasks
+                    </option>
+                    <option value="claude-fable-5">
+                      Claude Fable 5 — most capable, highest cost
                     </option>
                   </optgroup>
                   <optgroup label="OpenAI">

processing_scripts/llm_client/__init__.py

@@ -7,7 +7,7 @@
 -----
     from llm_client import create_audit_client, load_all_prompts, run_all
 
-    client = create_audit_client(model="claude-sonnet-4-6")   # or "gpt-4o"
+    client = create_audit_client(model="claude-sonnet-5")   # or "gpt-4o"
     prompts = load_all_prompts(PROMPTS_DIR)
     report  = run_all(client, prompts, all_slices)
 """

processing_scripts/llm_client/client.py

@@ -34,6 +34,25 @@ def is_gemini_model(model: str) -> bool:
     return model.startswith(_GEMINI_PREFIXES)
 
 
+# Claude models (Opus 4.6+, Sonnet 4.6+, Fable/Mythos 5) that reject
+# temperature/top_p/top_k (400 error) in favor of adaptive thinking + effort.
+_NO_SAMPLING_PARAMS_MODELS = {
+    "claude-fable-5",
+    "claude-mythos-5",
+    "claude-opus-5",
+    "claude-opus-4-8",
+    "claude-opus-4-7",
+    "claude-opus-4-6",
+    "claude-sonnet-5",
+    "claude-sonnet-4-6",
+}
+
+
+def supports_temperature(model: str) -> bool:
+    """Return False for Claude models that reject the `temperature` param."""
+    return model not in _NO_SAMPLING_PARAMS_MODELS
+
+
 class AuditClient:
     """
     Sends filled accessibility-audit prompts to the Claude API and returns
@@ -42,16 +61,21 @@ class AuditClient:
     Parameters
     ----------
     model : str
-        Claude model ID to use (default: claude-sonnet-4-6).
+        Claude model ID to use (default: claude-sonnet-5). Also accepts
+        ``claude-opus-5`` and ``claude-fable-5``, Anthropic's most capable
+        model. Newer Claude models (Opus 4.6+, Sonnet 4.6+, Fable/Mythos 5)
+        reject a ``temperature`` param entirely (see ``supports_temperature``),
+        which this client already accounts for.
     temperature : float
         Sampling temperature (default: 0.1 for consistent structured output).
+        Silently omitted from the request for models that reject it.
     max_tokens : int
         Maximum tokens to generate per response (default: 8192).
     """
 
     def __init__(
         self,
-        model: str = "claude-sonnet-4-6",
+        model: str = "claude-sonnet-5",
         temperature: float = 0.1,
         max_tokens: int = 8192,
     ):
@@ -82,12 +106,15 @@ def call(self, prompt_text: str, payload: dict) -> tuple[dict, dict]:
         """
         filled = prompt_text.replace("{payload}", json.dumps(payload, separators=(",", ":")))
 
-        message = self._client.messages.create(
-            model=self.model,
-            max_tokens=self.max_tokens,
-            temperature=self.temperature,
-            messages=[{"role": "user", "content": filled}],
-        )
+        request_kwargs = {
+            "model": self.model,
+            "max_tokens": self.max_tokens,
+            "messages": [{"role": "user", "content": filled}],
+        }
+        if supports_temperature(self.model):
+            request_kwargs["temperature"] = self.temperature
+
+        message = self._client.messages.create(**request_kwargs)
 
         usage = {
             "input_tokens": message.usage.input_tokens,
@@ -243,7 +270,7 @@ def call(self, prompt_text: str, payload: dict) -> tuple[dict, dict]:
 
 
 def create_audit_client(
-    model: str = "claude-sonnet-4-6",
+    model: str = "claude-sonnet-5",
     temperature: float = 0.1,
     max_tokens: int = 8192,
 ) -> AuditClient | OpenAIAuditClient | GeminiAuditClient: