feat(llm): bump seed-gen and patcher agents to newest models

Tier-preserving upgrade onto the models added in #544:
- seed-gen: primary claude-4.5-sonnet -> claude-4.6-sonnet; fallbacks
  collapsed to openai-gpt-5.5, gemini-pro (dropped redundant claude-4-sonnet
  now that claude-4.6-sonnet is primary).
- patcher (SWE, RootCause, Reflection, ContextRetriever, code-snippet
  helper): primary openai-gpt-4.1 -> openai-gpt-5.5; ContextRetriever
  cheap tier openai-gpt-4.1-mini -> openai-gpt-5.4-mini; fallback
  claude-4.5-sonnet -> claude-4.6-sonnet. gemini-pro left as-is.

Validated together with the litellm temperature fix in this PR via the
Docker e2e on example-libpng: seed-gen on claude-4.6-sonnet and patcher
on openai-gpt-5.5 both run healthy LLM turns with no errors.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

patcher/src/buttercup/patcher/agents/common.py

@@ -434,8 +434,8 @@ def _create_understand_code_snippet_chain() -> Runnable:
     return (
         UNDERSTAND_CODE_SNIPPET_PROMPT
         | create_default_llm_with_temperature(
-            model_name=ButtercupLLM.OPENAI_GPT_4_1.value,
-            fallback_models=[ButtercupLLM.CLAUDE_4_5_SONNET, ButtercupLLM.GEMINI_PRO],
+            model_name=ButtercupLLM.OPENAI_GPT_5_5.value,
+            fallback_models=[ButtercupLLM.CLAUDE_4_6_SONNET, ButtercupLLM.GEMINI_PRO],
         )
         | StrOutputParser()
     )

patcher/src/buttercup/patcher/agents/context_retriever.py

@@ -587,8 +587,8 @@ Remember to call the `test_instructions` tool to log and validate any test comma
 def _are_test_instructions_valid(instructions: str, output: bytes, error: bytes) -> bool:
     """Validate a set of test instructions by executing them inside the project environment."""
     llm = create_default_llm(
-        model_name=ButtercupLLM.OPENAI_GPT_4_1.value,
-        fallback_models=[ButtercupLLM.CLAUDE_4_5_SONNET, ButtercupLLM.GEMINI_PRO],
+        model_name=ButtercupLLM.OPENAI_GPT_5_5.value,
+        fallback_models=[ButtercupLLM.CLAUDE_4_6_SONNET, ButtercupLLM.GEMINI_PRO],
     )
     chain = ARE_VALID_TEST_INSTRUCTIONS_PROMPT | llm | StrOutputParser()
     res = chain.invoke(
@@ -717,10 +717,10 @@ class ContextRetrieverAgent(PatcherAgentBase):
 
     def __post_init__(self) -> None:
         """Initialize a few fields"""
-        self.llm = create_default_llm(model_name=ButtercupLLM.OPENAI_GPT_4_1.value)
-        self.cheap_llm = create_default_llm(model_name=ButtercupLLM.OPENAI_GPT_4_1_MINI.value)
+        self.llm = create_default_llm(model_name=ButtercupLLM.OPENAI_GPT_5_5.value)
+        self.cheap_llm = create_default_llm(model_name=ButtercupLLM.OPENAI_GPT_5_4_MINI.value)
         self.cheap_fallback_llms = [
-            create_default_llm(model_name=ButtercupLLM.CLAUDE_4_5_SONNET.value),
+            create_default_llm(model_name=ButtercupLLM.CLAUDE_4_6_SONNET.value),
             create_default_llm(model_name=ButtercupLLM.GEMINI_PRO.value),
         ]
 

patcher/src/buttercup/patcher/agents/reflection.py

@@ -405,9 +405,9 @@ class ReflectionAgent(PatcherAgentBase):
 
     def __post_init__(self) -> None:
         """Initialize a few fields"""
-        default_llm = create_default_llm(model_name=ButtercupLLM.OPENAI_GPT_4_1.value)
+        default_llm = create_default_llm(model_name=ButtercupLLM.OPENAI_GPT_5_5.value)
         fallback_llms: list[Runnable] = []
-        for fb_model in [ButtercupLLM.CLAUDE_4_5_SONNET, ButtercupLLM.GEMINI_PRO]:
+        for fb_model in [ButtercupLLM.CLAUDE_4_6_SONNET, ButtercupLLM.GEMINI_PRO]:
             fallback_llms.append(create_default_llm(model_name=fb_model.value))
 
         self.llm = default_llm.with_fallbacks(fallback_llms)

patcher/src/buttercup/patcher/agents/rootcause.py

@@ -130,7 +130,7 @@ class RootCauseAgent(PatcherAgentBase):
             "max_tokens": 20000,
         }
         default_llm = create_default_llm_with_temperature(
-            model_name=ButtercupLLM.OPENAI_GPT_4_1.value,
+            model_name=ButtercupLLM.OPENAI_GPT_5_5.value,
             **kwargs,
         )
         fallback_llms = [
@@ -138,7 +138,7 @@ class RootCauseAgent(PatcherAgentBase):
                 model_name=m.value,
                 **kwargs,
             )
-            for m in [ButtercupLLM.CLAUDE_4_5_SONNET, ButtercupLLM.GEMINI_PRO]
+            for m in [ButtercupLLM.CLAUDE_4_6_SONNET, ButtercupLLM.GEMINI_PRO]
         ]
         self.llm = default_llm.with_fallbacks(fallback_llms)
 

patcher/src/buttercup/patcher/agents/swe.py

@@ -347,12 +347,12 @@ class SWEAgent(PatcherAgentBase):
             "max_tokens": 20000,
         }
         self.default_llm = create_default_llm_with_temperature(
-            model_name=ButtercupLLM.OPENAI_GPT_4_1.value,
+            model_name=ButtercupLLM.OPENAI_GPT_5_5.value,
             **kwargs,
         )
         fallback_llms: list[Runnable] = []
         for fb_model in [
-            ButtercupLLM.CLAUDE_4_5_SONNET,
+            ButtercupLLM.CLAUDE_4_6_SONNET,
             ButtercupLLM.GEMINI_PRO,
         ]:
             fallback_llms.append(create_default_llm_with_temperature(model_name=fb_model.value, **kwargs))

seed-gen/src/buttercup/seed_gen/task.py

@@ -94,11 +94,10 @@ class Task:
 
     def __post_init__(self) -> None:
         fallbacks = [
-            ButtercupLLM.CLAUDE_4_SONNET,
-            ButtercupLLM.OPENAI_GPT_4_1,
+            ButtercupLLM.OPENAI_GPT_5_5,
             ButtercupLLM.GEMINI_PRO,
         ]
-        self.llm = Task.get_llm(ButtercupLLM.CLAUDE_4_5_SONNET, fallbacks)
+        self.llm = Task.get_llm(ButtercupLLM.CLAUDE_4_6_SONNET, fallbacks)
         self.tools = [
             get_function_definition,
             get_type_definition,