test(together_ai): select the live model from the cost map

`together_ai/openai/gpt-oss-20b` was hardcoded in two live tests and is no
longer served, so both failed on a vendor catalog change rather than on
anything litellm did.

Both call sites now resolve the cheapest non-deprecated together_ai chat
entry at runtime, filtered on the capabilities the tests actually exercise,
mirroring what tests/e2e/llm_translation/test_together_ai_e2e.py already
does. The selector lives in tests/_live_test_helpers.py so both lanes share
one implementation.

tests/_live_test_helpers.py

@@ -1,4 +1,7 @@
 import os
+from collections.abc import Mapping
+from datetime import date
+from typing import Any
 
 import pytest
 
@@ -8,3 +11,36 @@ def _skip_live_prompt_caching_test():
         pytest.skip("Live prompt-caching E2E tests are opt-in")
     if os.environ.get("CASSETTE_REDIS_URL"):
         pytest.skip("Live prompt-caching E2E tests cannot run under VCR replay")
+
+
+def cheapest_together_chat_model(*capability_flags: str) -> str:
+    import litellm
+
+    today = date.today().isoformat()
+
+    def qualifies(name: str, entry: Mapping[str, Any]) -> bool:
+        deprecation_date = entry.get("deprecation_date")
+        return (
+            name.startswith("together_ai/")
+            and entry.get("litellm_provider") == "together_ai"
+            and entry.get("mode") == "chat"
+            and (deprecation_date is None or deprecation_date > today)
+            and (entry.get("input_cost_per_token") or 0.0) > 0
+            and (entry.get("output_cost_per_token") or 0.0) > 0
+            and all(bool(entry.get(flag)) for flag in capability_flags)
+        )
+
+    candidates = sorted(
+        (
+            name
+            for name, entry in litellm.model_cost.items()
+            if isinstance(entry, Mapping) and qualifies(name, entry)
+        ),
+        key=lambda name: (
+            litellm.model_cost[name].get("input_cost_per_token") or 0.0,
+            litellm.model_cost[name].get("output_cost_per_token") or 0.0,
+            name,
+        ),
+    )
+    assert candidates, f"no live together_ai chat model in the cost map satisfies {capability_flags}"
+    return candidates[0]

tests/llm_translation/test_together_ai.py

@@ -3,6 +3,7 @@ Test TogetherAI LLM
 """
 
 from base_llm_unit_tests import BaseLLMChatTest
+from tests._live_test_helpers import cheapest_together_chat_model
 import json
 import os
 from datetime import datetime
@@ -16,7 +17,11 @@ import pytest
 class TestTogetherAI(BaseLLMChatTest):
     def get_base_completion_call_args(self) -> dict:
         litellm.set_verbose = True
-        return {"model": "together_ai/openai/gpt-oss-20b"}
+        return {
+            "model": cheapest_together_chat_model(
+                "supports_function_calling", "supports_response_schema"
+            )
+        }
 
     def test_tool_call_no_arguments(self, tool_call_no_arguments):
         """Test that tool calls with no arguments is translated correctly. Relevant issue: https://github.com/BerriAI/litellm/issues/6833"""

tests/local_testing/test_text_completion.py

@@ -1,7 +1,11 @@
 import asyncio
 import json
+import os
+import sys
 import traceback
 
+sys.path.insert(0, os.path.abspath(os.path.join(os.path.dirname(__file__), "..", "..")))
+
 from dotenv import load_dotenv
 
 load_dotenv()
@@ -12,6 +16,7 @@ from unittest.mock import MagicMock, patch
 import pytest
 
 import litellm
+from tests._live_test_helpers import cheapest_together_chat_model
 from litellm import (
     RateLimitError,
     TextCompletionResponse,
@@ -4030,7 +4035,7 @@ def test_async_text_completion_together_ai():
     async def test_get_response():
         try:
             response = await litellm.atext_completion(
-                model="together_ai/openai/gpt-oss-20b",
+                model=cheapest_together_chat_model(),
                 prompt="good morning",
                 max_tokens=10,
             )