diff --git a/agent/auxiliary_client.py b/agent/auxiliary_client.py
index 44b11dfaa6d0..4bd1710ef252 100644
--- a/agent/auxiliary_client.py
+++ b/agent/auxiliary_client.py
@@ -4726,7 +4726,9 @@ def _build_call_kwargs(
 
     if max_tokens is not None:
         # Codex adapter handles max_tokens internally; OpenRouter/Nous use max_tokens.
-        # Direct OpenAI api.openai.com with newer models needs max_completion_tokens.
+        # Direct OpenAI api.openai.com and GitHub Copilot api.githubcopilot.com with
+        # newer GPT-4o/o-series/GPT-5 models reject max_tokens and require
+        # max_completion_tokens instead.
         # ZAI vision models (glm-4v-flash, glm-4v-plus, etc.) reject max_tokens with
         # error code 1210 ("API 调用参数有误") on multimodal requests — skip it.
         _model_lower = (model or "").lower()
@@ -4734,14 +4736,22 @@ def _build_call_kwargs(
             provider == "zai"
             and ("4v" in _model_lower or "5v" in _model_lower or "-v" in _model_lower)
         )
+        # Endpoints that require max_completion_tokens for newer OpenAI models,
+        # keyed on the resolved request hostname rather than the provider label.
+        # GitHub Copilot normalises to provider="copilot" (not "custom"), so a
+        # provider=="custom" gate alone misses it — mirror auxiliary_max_tokens_param,
+        # which already lists both hosts. See issue #34530.
+        _resolved_base = base_url
+        if not _resolved_base and provider == "custom":
+            _resolved_base = _current_custom_base_url()
+        _wants_max_completion = base_url_hostname(_resolved_base) in {
+            "api.openai.com",
+            "api.githubcopilot.com",
+        }
         if _skip_max_tokens:
             pass  # ZAI vision models do not accept max_tokens
-        elif provider == "custom":
-            custom_base = base_url or _current_custom_base_url()
-            if base_url_hostname(custom_base) == "api.openai.com":
-                kwargs["max_completion_tokens"] = max_tokens
-            else:
-                kwargs["max_tokens"] = max_tokens
+        elif _wants_max_completion:
+            kwargs["max_completion_tokens"] = max_tokens
         else:
             kwargs["max_tokens"] = max_tokens
 
diff --git a/tests/agent/test_auxiliary_client.py b/tests/agent/test_auxiliary_client.py
index 66e52b6c10b9..d6e684c48fd2 100644
--- a/tests/agent/test_auxiliary_client.py
+++ b/tests/agent/test_auxiliary_client.py
@@ -88,6 +88,87 @@ def test_uses_max_completion_tokens_for_github_copilot_custom_base_path(self):
             assert auxiliary_max_tokens_param(2048) == {"max_completion_tokens": 2048}
 
 
+class TestBuildCallKwargsMaxTokensParam:
+    """_build_call_kwargs must emit max_completion_tokens for endpoints that
+    reject max_tokens on newer OpenAI models, keyed on the resolved request
+    hostname rather than the provider label.
+
+    Regression test for issue #34530: auxiliary context compression configured
+    with provider=github-copilot (which normalises to provider="copilot", not
+    "custom") and a GPT-5 model sent max_tokens and failed with HTTP 400
+    "Unsupported parameter: 'max_tokens' ... Use 'max_completion_tokens'".
+    """
+
+    def test_github_copilot_provider_uses_max_completion_tokens(self):
+        # github-copilot normalises to provider="copilot"; the bug was that the
+        # max_completion_tokens path was gated on provider=="custom".
+        kwargs = _build_call_kwargs(
+            provider="copilot",
+            model="gpt-5.4",
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=1024,
+            base_url="https://api.githubcopilot.com",
+        )
+        assert kwargs.get("max_completion_tokens") == 1024
+        assert "max_tokens" not in kwargs
+
+    def test_github_copilot_base_with_path_uses_max_completion_tokens(self):
+        kwargs = _build_call_kwargs(
+            provider="copilot",
+            model="gpt-5.5",
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=512,
+            base_url="https://api.githubcopilot.com/chat/completions",
+        )
+        assert kwargs.get("max_completion_tokens") == 512
+        assert "max_tokens" not in kwargs
+
+    def test_openai_base_still_uses_max_completion_tokens(self):
+        kwargs = _build_call_kwargs(
+            provider="openai",
+            model="gpt-5.4",
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=2048,
+            base_url="https://api.openai.com/v1",
+        )
+        assert kwargs.get("max_completion_tokens") == 2048
+        assert "max_tokens" not in kwargs
+
+    def test_custom_openai_base_still_uses_max_completion_tokens(self):
+        # Preserve the pre-existing provider=="custom" + api.openai.com behaviour.
+        kwargs = _build_call_kwargs(
+            provider="custom",
+            model="gpt-5.4",
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=2048,
+            base_url="https://api.openai.com/v1",
+        )
+        assert kwargs.get("max_completion_tokens") == 2048
+        assert "max_tokens" not in kwargs
+
+    def test_openrouter_still_uses_max_tokens(self):
+        kwargs = _build_call_kwargs(
+            provider="openrouter",
+            model="openai/gpt-5.4",
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=4096,
+            base_url="https://openrouter.ai/api/v1",
+        )
+        assert kwargs.get("max_tokens") == 4096
+        assert "max_completion_tokens" not in kwargs
+
+    def test_no_base_url_non_custom_provider_uses_max_tokens(self):
+        # Defensive: an unknown host (empty) must keep max_tokens, not flip.
+        kwargs = _build_call_kwargs(
+            provider="deepseek",
+            model="deepseek-chat",
+            messages=[{"role": "user", "content": "hi"}],
+            max_tokens=1000,
+        )
+        assert kwargs.get("max_tokens") == 1000
+        assert "max_completion_tokens" not in kwargs
+
+
 class TestNormalizeAuxProvider:
     def test_maps_github_copilot_aliases(self):
         assert _normalize_aux_provider("github") == "copilot"
diff --git a/tests/agent/test_unsupported_temperature_retry.py b/tests/agent/test_unsupported_temperature_retry.py
index 82d8d3208d29..3bffb1b4062a 100644
--- a/tests/agent/test_unsupported_temperature_retry.py
+++ b/tests/agent/test_unsupported_temperature_retry.py
@@ -112,8 +112,12 @@ def test_retries_once_without_temperature(self, error_message):
         retry_kwargs = client.chat.completions.create.call_args_list[1].kwargs
         assert first_kwargs["temperature"] == 0.3
         assert "temperature" not in retry_kwargs
-        # other kwargs preserved
-        assert retry_kwargs["max_tokens"] == 500
+        # other kwargs preserved. api.openai.com normalises max_tokens to
+        # max_completion_tokens (issue #34530), so accept either.
+        assert (
+            retry_kwargs.get("max_tokens") == 500
+            or retry_kwargs.get("max_completion_tokens") == 500
+        )
 
     def test_non_temperature_400_does_not_retry_as_temperature(self):
         """Unrelated 400s (e.g. bad tool role) must not silently drop temp."""
@@ -207,7 +211,12 @@ async def test_async_retries_once_without_temperature(self):
         retry_kwargs = client.chat.completions.create.call_args_list[1].kwargs
         assert first_kwargs["temperature"] == 0.3
         assert "temperature" not in retry_kwargs
-        assert retry_kwargs["max_tokens"] == 500
+        # api.openai.com normalises max_tokens to max_completion_tokens
+        # (issue #34530), so accept either form on the retry call.
+        assert (
+            retry_kwargs.get("max_tokens") == 500
+            or retry_kwargs.get("max_completion_tokens") == 500
+        )
 
     @pytest.mark.asyncio
     async def test_async_non_temperature_400_does_not_retry(self):