From 84709a00d979c396dfb34f5d60b2992efb5a9284 Mon Sep 17 00:00:00 2001
From: Husam <husamemad60@gmail.com>
Date: Thu, 30 Jul 2026 13:30:00 +0300
Subject: [PATCH] fix(llm): omit temperature for major-only Opus ids
 (claude-opus-5) (#5761)

The version pattern in _anthropic_rejects_temperature() required a minor
component, so major-only ids like `claude-opus-5` never matched and the
guard reported that the model accepts `temperature`. Anthropic rejects the
field outright on Opus 4.7+, so every such call returned HTTP 400 and the
stream aborted with zero tokens ("the model returned an empty response").

Make the minor optional and read a missing minor as `.0`. The major is also
capped at 1-2 digits with a no-trailing-digit lookahead, mirroring the
minor: once the minor is optional, a greedy major would swallow the date in
`claude-3-opus-20240229` and read it as version 20240229, dropping
temperature from a model that accepts it.

Fixes #5753

Co-authored-by: Alexandre Teixeira <111787685+alteixeira20@users.noreply.github.com>
---
 src/llm_core.py                            | 26 ++++++++++++++++------
 tests/test_llm_core_anthropic_temp_omit.py | 26 +++++++++++++++++++++-
 2 files changed, 44 insertions(+), 8 deletions(-)

diff --git a/src/llm_core.py b/src/llm_core.py
index 4dec32376..3e84c1060 100644
--- a/src/llm_core.py
+++ b/src/llm_core.py
@@ -1237,15 +1237,27 @@ def _anthropic_rejects_temperature(model: str) -> bool:
         return False
     # `(?<![a-z])` anchors "opus" to a word boundary so a substring match like
     # `oct-opus`/`octopus-4-8` can't be read as Opus (it would otherwise strip
-    # temperature). Cap the minor at 1-2 digits and forbid a trailing digit so a
-    # dated id like `claude-opus-4-20250514` (Opus 4.0) parses as major-only (no
-    # minor match, kept) instead of reading the date `20250514` as a giant minor
-    # that would falsely test >= 4.7. Dated 4.7+ snapshots (`claude-opus-4-7-
-    # 20260201`) keep their explicit minor and are still matched.
-    match = re.search(r"(?<![a-z])opus[-_]?(\d+)[-_.](\d{1,2})(?!\d)", model.lower())
+    # temperature). Both version components are capped at 1-2 digits and forbid a
+    # trailing digit, so an 8-digit date can never be read as a version number:
+    # `claude-opus-4-20250514` (Opus 4.0) parses as major-only rather than reading
+    # `20250514` as a giant minor, and `claude-3-opus-20240229` (legacy Claude 3
+    # Opus, date directly after "opus-") fails to match at all rather than reading
+    # the date as a giant major. Dated 4.7+ snapshots (`claude-opus-4-7-20260201`)
+    # keep their explicit minor and are still matched.
+    #
+    # The minor is optional and a missing minor reads as `.0`, so major-only ids
+    # like `claude-opus-5` are correctly treated as >= 4.7 (issue #5753). Without
+    # this, every Opus 5 call kept `temperature` and failed with HTTP 400 — visible
+    # only on paths that pass a temperature, e.g. scheduled tasks inheriting
+    # `stream_agent_loop`'s 0.3 default, which returned empty responses.
+    match = re.search(
+        r"(?<![a-z])opus[-_]?(\d{1,2})(?!\d)(?:[-_.](\d{1,2})(?!\d))?", model.lower()
+    )
     if not match:
         return False
-    return (int(match.group(1)), int(match.group(2))) >= (4, 7)
+    major = int(match.group(1))
+    minor = int(match.group(2)) if match.group(2) else 0
+    return (major, minor) >= (4, 7)
 
 # Reasoning effort level sent to Mistral thinking-capable models. Mistral's
 # API accepts "high", "medium", "low", "none" — see
diff --git a/tests/test_llm_core_anthropic_temp_omit.py b/tests/test_llm_core_anthropic_temp_omit.py
index 2274f1dc9..f7d26aef0 100644
--- a/tests/test_llm_core_anthropic_temp_omit.py
+++ b/tests/test_llm_core_anthropic_temp_omit.py
@@ -29,6 +29,13 @@ from src.llm_core import _anthropic_rejects_temperature, _build_anthropic_payloa
         "anthropic/claude-opus-4-7",  # tolerate a provider-prefixed id
         "claude-opus-4-10",  # future minor still >= 4.7
         "claude-opus-5-0",  # future major
+        # Major-only ids: a missing minor reads as `.0`, so these are >= 4.7 too
+        # (issue #5753). Before the fix the version pattern required a minor, so
+        # these fell through to "accepts temperature" and every call 400'd.
+        "claude-opus-5",
+        "claude-opus-5-20260101",  # major-only + dated snapshot
+        "anthropic/claude-opus-5",  # major-only behind a provider prefix
+        "claude-opus-6",  # future major-only
     ],
 )
 def test_opus_47_plus_rejects_temperature(model):
@@ -48,7 +55,10 @@ def test_opus_47_plus_rejects_temperature(model):
         "claude-opus-4-6-20251201",  # dated 4.6 snapshot — older, still keeps temperature
         "claude-sonnet-4-6",
         "claude-3-5-sonnet",
-        "claude-3-opus-20240229",  # legacy Claude 3 Opus — no opus-N-M pattern, kept
+        "claude-3-opus-20240229",  # legacy Claude 3 Opus — date directly after
+        # "opus-", so the major must not swallow it as version 20240229 (that is
+        # what makes capping the major at 1-2 digits necessary once the minor
+        # became optional in #5753).
         "claude-haiku-4-5",
         "claude-x",
         "octopus-4-8",  # "opus" only as a substring of another word — must not match
@@ -87,6 +97,20 @@ def test_payload_keeps_temperature_for_older_models():
     assert _payload("claude-3-5-sonnet", 1.2)["temperature"] == 1.0
 
 
+def test_payload_omits_temperature_for_major_only_opus_5():
+    # Issue #5753: the scheduled-task path calls stream_agent_loop() without a
+    # temperature and inherits its 0.3 default, so `claude-opus-5` 400'd on every
+    # run and surfaced as "the model returned an empty response". Interactive chat
+    # leaves temperature None and never hit it.
+    assert "temperature" not in _payload("claude-opus-5", 0.3)
+
+
+def test_payload_keeps_temperature_for_legacy_claude_3_opus():
+    # Guards the major-digit cap: `opus-20240229` must not parse as version
+    # 20240229, or Claude 3 Opus would silently lose the caller's temperature.
+    assert _payload("claude-3-opus-20240229", 0.5)["temperature"] == 0.5
+
+
 def test_payload_keeps_temperature_for_dated_opus_4_0():
     # Anthropic's dated id for Opus 4.0 (claude-opus-4-20250514) is in this repo's
     # ANTHROPIC_MODELS list. The date must not be misread as a >= 4.7 minor, or the