seb/hermes-agent
1
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

fix(kimi): send thinking xor reasoning_effort, never both

Browse Source 
The standalone Kimi/Moonshot profile (api.moonshot.ai/v1) sent both
extra_body.thinking AND a top-level reasoning_effort. With no reasoning
config it even defaulted to thinking:enabled + reasoning_effort:medium,
pairing them on every default call. Moonshot treats these as mutually
exclusive (cannot specify both 'thinking' and 'reasoning_effort').

Align with the kimi-k2 handling already shipped for the opencode-go relay:
send effort when a recognized low|medium|high is requested, otherwise fall
back to the extra_body.thinking toggle. Disabled sends thinking:disabled
only. Never both.

Reported by Cars29 (NOUS Discord). DeepSeek was deliberately left untouched:
its native endpoint accepts both (verified by the live guardrail in
test_deepseek_v4_thinking_live.py), so the report's DeepSeek claim does not
hold there.

Tests: tests/plugins/model_providers/test_kimi_profile.py pins the xor
contract across all config shapes.
This commit is contained in:
teknium1
2026-06-07 01:24:29 -07:00
committed by Teknium
parent 03392b67d6
commit ce4e74b350
2 changed files with 146 additions and 7 deletions
Download Patch File Download Diff File Expand all files Collapse all files
plugins/model-providers/kimi-coding/__init__.py
+16 -7
View File
@@ -14,19 +14,28 @@ from providers.base import OMIT_TEMPERATURE, ProviderProfile
class KimiProfile(ProviderProfile):
"""Kimi/Moonshot — temperature omitted, thinking + reasoning_effort."""
"""Kimi/Moonshot — temperature omitted, thinking xor reasoning_effort."""
def build_api_kwargs_extras(
self, *, reasoning_config: dict | None = None, **context
) -> tuple[dict[str, Any], dict[str, Any]]:
"""Kimi uses extra_body.thinking + top-level reasoning_effort."""
"""Kimi reasoning controls.
Moonshot's wire shape treats ``extra_body.thinking`` (a binary toggle)
and a top-level ``reasoning_effort`` as mutually exclusive — sending
both is at best redundant and risks "cannot specify both 'thinking' and
'reasoning_effort'" (HTTP 400). This mirrors the kimi-k2 handling on the
opencode-go relay: send effort when one is requested, otherwise fall
back to ``extra_body.thinking`` — never both.
"""
extra_body = {}
top_level = {}
if not reasoning_config or not isinstance(reasoning_config, dict):
# No config → thinking enabled, default effort
# No config → thinking enabled, let the server pick the depth.
# (Previously also sent reasoning_effort="medium", which paired
# thinking + effort on every default call.)
extra_body["thinking"] = {"type": "enabled"}
top_level["reasoning_effort"] = "medium"
return extra_body, top_level
enabled = reasoning_config.get("enabled", True)
@@ -34,13 +43,13 @@ class KimiProfile(ProviderProfile):
extra_body["thinking"] = {"type": "disabled"}
return extra_body, top_level
# Enabled
extra_body["thinking"] = {"type": "enabled"}
# Enabled: prefer an explicit effort; only fall back to extra_body
# thinking when no recognized effort is requested.
effort = (reasoning_config.get("effort") or "").strip().lower()
if effort in {"low", "medium", "high"}:
top_level["reasoning_effort"] = effort
else:
top_level["reasoning_effort"] = "medium"
extra_body["thinking"] = {"type": "enabled"}
return extra_body, top_level