Fix Claude Sonnet 5 & Opus 4.8 compatibility

gui_agents/s1/mllm/MultimodalEngine.py

@@ -11,6 +11,10 @@
 import openai
 import requests
 from anthropic import Anthropic
+from gui_agents.utils import (
+    anthropic_supports_temperature,
+    extract_anthropic_text,
+)
 from openai import APIConnectionError, APIError, AzureOpenAI, OpenAI, RateLimitError
 from PIL import Image
 
@@ -115,18 +119,16 @@ def __init__(self, api_key=None, model=None, **kwargs):
     )
     def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
         """Generate the next message based on previous messages"""
-        return (
-            self.llm_client.messages.create(
-                system=messages[0]["content"][0]["text"],
-                model=self.model,
-                messages=messages[1:],
-                max_tokens=max_new_tokens if max_new_tokens else 4096,
-                temperature=temperature,
-                **kwargs,
-            )
-            .content[0]
-            .text
-        )
+        request_kwargs = {
+            "system": messages[0]["content"][0]["text"],
+            "model": self.model,
+            "messages": messages[1:],
+            "max_tokens": max_new_tokens if max_new_tokens else 4096,
+            **kwargs,
+        }
+        if anthropic_supports_temperature(self.model):
+            request_kwargs["temperature"] = temperature
+        return extract_anthropic_text(self.llm_client.messages.create(**request_kwargs))
 
 
 class OpenAIEmbeddingEngine(LMMEngine):

gui_agents/s2/core/engine.py

@@ -3,6 +3,11 @@
 import backoff
 import numpy as np
 from anthropic import Anthropic
+from gui_agents.utils import (
+    anthropic_supports_temperature,
+    extract_anthropic_text,
+    extract_anthropic_thinking,
+)
 from openai import (
     AzureOpenAI,
     APIConnectionError,
@@ -215,21 +220,19 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
                 thinking={"type": "enabled", "budget_tokens": 4096},
                 **kwargs,
             )
-            thoughts = full_response.content[0].thinking
-            print("CLAUDE 3.7 THOUGHTS:", thoughts)
-            return full_response.content[1].text
-        return (
-            self.llm_client.messages.create(
-                system=messages[0]["content"][0]["text"],
-                model=self.model,
-                messages=messages[1:],
-                max_tokens=max_new_tokens if max_new_tokens else 4096,
-                temperature=temperature,
-                **kwargs,
-            )
-            .content[0]
-            .text
-        )
+            thoughts = extract_anthropic_thinking(full_response)
+            print("CLAUDE THOUGHTS:", thoughts)
+            return extract_anthropic_text(full_response)
+        request_kwargs = {
+            "system": messages[0]["content"][0]["text"],
+            "model": self.model,
+            "messages": messages[1:],
+            "max_tokens": max_new_tokens if max_new_tokens else 4096,
+            **kwargs,
+        }
+        if anthropic_supports_temperature(self.model):
+            request_kwargs["temperature"] = temperature
+        return extract_anthropic_text(self.llm_client.messages.create(**request_kwargs))
 
 
 class LMMEngineGemini(LMMEngine):

gui_agents/s2_5/core/engine.py

@@ -2,6 +2,11 @@
 
 import backoff
 from anthropic import Anthropic
+from gui_agents.utils import (
+    anthropic_supports_temperature,
+    extract_anthropic_text,
+    extract_anthropic_thinking,
+)
 from openai import (
     AzureOpenAI,
     APIConnectionError,
@@ -107,20 +112,17 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
                 thinking={"type": "enabled", "budget_tokens": 4096},
                 **kwargs,
             )
-            thoughts = full_response.content[0].thinking
-            return full_response.content[1].text
-        return (
-            self.llm_client.messages.create(
-                system=messages[0]["content"][0]["text"],
-                model=self.model,
-                messages=messages[1:],
-                max_tokens=max_new_tokens if max_new_tokens else 4096,
-                temperature=temp,
-                **kwargs,
-            )
-            .content[0]
-            .text
-        )
+            return extract_anthropic_text(full_response)
+        request_kwargs = {
+            "system": messages[0]["content"][0]["text"],
+            "model": self.model,
+            "messages": messages[1:],
+            "max_tokens": max_new_tokens if max_new_tokens else 4096,
+            **kwargs,
+        }
+        if anthropic_supports_temperature(self.model):
+            request_kwargs["temperature"] = temp
+        return extract_anthropic_text(self.llm_client.messages.create(**request_kwargs))
 
     @backoff.on_exception(
         backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
@@ -140,8 +142,8 @@ def generate_with_thinking(
             **kwargs,
         )
 
-        thoughts = full_response.content[0].thinking
-        answer = full_response.content[1].text
+        thoughts = extract_anthropic_thinking(full_response)
+        answer = extract_anthropic_text(full_response)
         full_response = (
             f"<thoughts>\n{thoughts}\n</thoughts>\n\n<answer>\n{answer}\n</answer>\n"
         )

gui_agents/s3/agents/worker.py

@@ -46,7 +46,7 @@ def __init__(
         """
         super().__init__(worker_engine_params, platform)
 
-        self.temperature = worker_engine_params.get("temperature", 0.0)
+        self.temperature = worker_engine_params.get("temperature") or 0.0
         self.use_thinking = worker_engine_params.get("model", "") in [
             "claude-opus-4-20250514",
             "claude-sonnet-4-20250514",

gui_agents/s3/cli_app.py

@@ -335,8 +335,9 @@ def main():
         "model": args.model,
         "base_url": args.model_url,
         "api_key": args.model_api_key,
-        "temperature": getattr(args, "model_temperature", None),
     }
+    if args.model_temperature is not None:
+        engine_params["temperature"] = args.model_temperature
 
     # Load the grounding engine from a custom endpoint
     engine_params_for_grounding = {

gui_agents/s3/core/engine.py

@@ -2,6 +2,11 @@
 
 import backoff
 from anthropic import Anthropic
+from gui_agents.utils import (
+    anthropic_supports_temperature,
+    extract_anthropic_text,
+    extract_anthropic_thinking,
+)
 from openai import (
     AzureOpenAI,
     APIConnectionError,
@@ -95,8 +100,13 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
                 "An API Key needs to be provided in either the api_key parameter or as an environment variable named ANTHROPIC_API_KEY"
             )
         self.llm_client = Anthropic(api_key=api_key)
-        # Use the instance temperature if not specified in the call
-        temp = self.temperature if temperature is None else temperature
+        # Prefer instance temperature when set; otherwise use call arg or default to 0.0
+        if self.temperature is not None:
+            temp = self.temperature
+        elif temperature is not None:
+            temp = temperature
+        else:
+            temp = 0.0
         if self.thinking:
             full_response = self.llm_client.messages.create(
                 system=messages[0]["content"][0]["text"],
@@ -106,20 +116,17 @@ def generate(self, messages, temperature=0.0, max_new_tokens=None, **kwargs):
                 thinking={"type": "enabled", "budget_tokens": 4096},
                 **kwargs,
             )
-            thoughts = full_response.content[0].thinking
-            return full_response.content[1].text
-        return (
-            self.llm_client.messages.create(
-                system=messages[0]["content"][0]["text"],
-                model=self.model,
-                messages=messages[1:],
-                max_tokens=max_new_tokens if max_new_tokens else 4096,
-                temperature=temp,
-                **kwargs,
-            )
-            .content[0]
-            .text
-        )
+            return extract_anthropic_text(full_response)
+        request_kwargs = {
+            "system": messages[0]["content"][0]["text"],
+            "model": self.model,
+            "messages": messages[1:],
+            "max_tokens": max_new_tokens if max_new_tokens else 4096,
+            **kwargs,
+        }
+        if anthropic_supports_temperature(self.model):
+            request_kwargs["temperature"] = temp
+        return extract_anthropic_text(self.llm_client.messages.create(**request_kwargs))
 
     @backoff.on_exception(
         backoff.expo, (APIConnectionError, APIError, RateLimitError), max_time=60
@@ -144,8 +151,8 @@ def generate_with_thinking(
             **kwargs,
         )
 
-        thoughts = full_response.content[0].thinking
-        answer = full_response.content[1].text
+        thoughts = extract_anthropic_thinking(full_response)
+        answer = extract_anthropic_text(full_response)
         full_response = (
             f"<thoughts>\n{thoughts}\n</thoughts>\n\n<answer>\n{answer}\n</answer>\n"
         )

gui_agents/utils.py

@@ -1,11 +1,52 @@
 """General utility."""
 
 import platform
+import re
 import requests
 import zipfile
 import io
 import os
 
+# Anthropic models that reject non-default sampling params (temperature, top_p, top_k).
+_ANTHROPIC_NO_TEMPERATURE_PATTERNS = (
+    re.compile(r"sonnet-5"),
+    re.compile(r"opus-4-7"),
+    re.compile(r"opus-4-8"),
+)
+
+
+def anthropic_supports_temperature(model: str) -> bool:
+    """Return False for Anthropic models that reject the temperature parameter."""
+    model_lower = model.lower()
+    return not any(
+        pattern.search(model_lower) for pattern in _ANTHROPIC_NO_TEMPERATURE_PATTERNS
+    )
+
+
+def extract_anthropic_text(response) -> str:
+    """Concatenate text from an Anthropic Messages response.
+
+    Newer models (e.g. Sonnet 5) enable adaptive thinking by default, so the
+    response content can contain ThinkingBlocks that have no `text` attribute.
+    This skips non-text blocks and returns only the assistant's text output.
+    """
+    texts = [
+        block.text
+        for block in response.content
+        if getattr(block, "type", None) == "text"
+    ]
+    return "".join(texts)
+
+
+def extract_anthropic_thinking(response) -> str:
+    """Concatenate thinking/reasoning text from an Anthropic Messages response."""
+    thoughts = [
+        getattr(block, "thinking", "")
+        for block in response.content
+        if getattr(block, "type", None) == "thinking"
+    ]
+    return "".join(thoughts)
+
 
 def download_kb_data(
     version="s2",