personal memory agent
Fix Gemini JSON output tests hitting token limit
Gemini 3's implicit thinking consumes part of max_output_tokens budget.
With 100 tokens, ~92 went to thinking leaving insufficient tokens for
JSON output, causing MAX_TOKENS finish reason. Increased to 500 tokens.
🤖 Generated with [Claude Code](https://claude.com/claude-code)
Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
+4
-2
+4
-2
tests/integration/test_gemini_generate.py
+4
-2
tests/integration/test_gemini_generate.py
···
102
102
os.environ["JOURNAL_PATH"] = journal_path
103
103
104
104
# Test JSON output mode
105
+
# Note: Gemini 3 uses implicit thinking which consumes part of max_output_tokens
105
106
response = gemini_generate(
106
107
"Create a JSON object with fields 'name' (value: 'test') and 'number' (value: 42)",
107
108
model=GEMINI_FLASH,
108
109
json_output=True,
109
110
temperature=0.1,
110
-
max_output_tokens=100,
111
+
max_output_tokens=500,
111
112
)
112
113
113
114
assert response is not None
···
420
421
os.environ["JOURNAL_PATH"] = journal_path
421
422
422
423
# Test async JSON output mode
424
+
# Note: Gemini 3 uses implicit thinking which consumes part of max_output_tokens
423
425
response = await gemini_agenerate(
424
426
"Create a JSON object with fields 'status' (value: 'success') and 'value' (value: 100)",
425
427
model=GEMINI_FLASH,
426
428
json_output=True,
427
429
temperature=0.1,
428
-
max_output_tokens=100,
430
+
max_output_tokens=500,
429
431
)
430
432
431
433
assert response is not None