fix(reasoning): add deepseek-v4-flash and deepseek-v4-pro to reasoning timeout floor

DeepSeek V4 models (deepseek-v4-flash, deepseek-v4-pro) emit
reasoning_content in a separate delta field before final content,
requiring the same 600s stale timeout floor as R1. Without this,
streams hang for 30–50s with APITimeoutError on providers like
opencode-go while direct calls succeed in ~3s.

Fixes #60338.

agent/reasoning_timeouts.py

@@ -66,9 +66,13 @@ _REASONING_STALE_TIMEOUT_FLOORS: tuple[tuple[str, int], ...] = (
     ("nemotron-3-ultra", 600),
     ("nemotron-3-super", 600),
     ("nemotron-3-nano",  300),
-    # DeepSeek — R1 reasoning model on hosted NIM / DeepSeek direct.
+    # DeepSeek — R1 and V4 reasoning models on hosted NIM / DeepSeek direct.
+    # V4 series emits reasoning_content in a separate delta field before
+    # final content, requiring the same extended stale timeout floor.
     ("deepseek-r1", 600),
     ("deepseek-reasoner", 600),
+    ("deepseek-v4-flash", 600),
+    ("deepseek-v4-pro", 600),
     # Qwen — QwQ reasoning + Qwen3 thinking variants.  QwQ-32B
     # preview is the stable slug; ``qwen3`` covers the family of
     # thinking-mode Qwen3 models (qwen3-235b-a22b, qwen3-32b, etc.)
@@ -190,6 +194,10 @@ def get_reasoning_stale_timeout_floor(model: object) -> Optional[float]:
     300.0
     >>> get_reasoning_stale_timeout_floor("deepseek/deepseek-r1")
     600.0
+    >>> get_reasoning_stale_timeout_floor("deepseek/deepseek-v4-flash")
+    600.0
+    >>> get_reasoning_stale_timeout_floor("deepseek/deepseek-v4-pro")
+    600.0
     >>> get_reasoning_stale_timeout_floor("qwen/qwen3-235b-a22b-thinking")
     180.0
     >>> get_reasoning_stale_timeout_floor("x-ai/grok-4-fast-reasoning")

tests/agent/test_reasoning_timeouts.py

@@ -0,0 +1,46 @@
+"""Test get_reasoning_stale_timeout_floor for DeepSeek V4 reasoning models."""
+
+from agent.reasoning_timeouts import get_reasoning_stale_timeout_floor
+
+
+def test_deepseek_v4_reasoning_models_have_timeout_floor():
+    """DeepSeek V4 reasoning models should have 600s timeout floor.
+
+    DeepSeek V4 models (deepseek-v4-flash, deepseek-v4-pro) emit
+    reasoning_content in a separate delta field before final content,
+    requiring the same extended stale timeout floor as R1.
+    See #60338.
+    """
+    # Direct model names (no aggregator prefix)
+    assert get_reasoning_stale_timeout_floor("deepseek-v4-flash") == 600.0
+    assert get_reasoning_stale_timeout_floor("deepseek-v4-pro") == 600.0
+
+    # With aggregator prefixes (common usage)
+    assert get_reasoning_stale_timeout_floor("deepseek/deepseek-v4-flash") == 600.0
+    assert get_reasoning_stale_timeout_floor("deepseek/deepseek-v4-pro") == 600.0
+
+    # With custom provider prefixes
+    assert get_reasoning_stale_timeout_floor("opencode/deepseek-v4-flash") == 600.0
+    assert get_reasoning_stale_timeout_floor("custom/deepseek-v4-pro") == 600.0
+
+
+def test_deepseek_v4_timeout_floor_matches_r1():
+    """V4 models should have the same floor as R1 (600s)."""
+    v4_flash_floor = get_reasoning_stale_timeout_floor("deepseek-v4-flash")
+    v4_pro_floor = get_reasoning_stale_timeout_floor("deepseek-v4-pro")
+    r1_floor = get_reasoning_stale_timeout_floor("deepseek-r1")
+    reasoner_floor = get_reasoning_stale_timeout_floor("deepseek-reasoner")
+
+    assert v4_flash_floor == v4_pro_floor == r1_floor == reasoner_floor == 600.0
+
+
+def test_deepseek_v4_does_not_match_non_reasoning_deepseek():
+    """Non-reasoning deepseek variants should not match V4 patterns."""
+    # deepseek-chat is not a reasoning model and should not match
+    # the deepseek-v4 patterns (requires start-of-slug anchor).
+    assert get_reasoning_stale_timeout_floor("deepseek-chat") is None
+    assert get_reasoning_stale_timeout_floor("deepseek/deepseek-chat") is None
+
+    # But V4 patterns should still match V4 models even with non-V4 prefixes
+    assert get_reasoning_stale_timeout_floor("some-deepseek-v4-flash") is None  # wrong prefix
+    assert get_reasoning_stale_timeout_floor("deepseek-v4-flash-model") == 600.0  # suffix OK