Gemini 3.1 Flash Lite: cheap, quick, still not closed
Gemini 3.1 Flash Lite ran the full current suite quickly and cheaply. It reached five core passes, generated a valid City Plan SVG, and still resolved none of the nine practical paperwork cases strictly.
Gemini 3.1 Flash Lite is included as a cheap API comparison, not as a local model. It is useful because it shows what a very accessible current API model does on the same paperwork contracts as the Mac mini runs.
The short version is familiar: it understood enough to produce several near misses, but the exact job did not close. Proof codes, evidence paths, attachment indexes, normalized text, and payment reconciliation still broke the strict oracle.
Where it worked
- Fast responses across the suite.
- Reached core-pass level on five of nine practical cases.
- Produced visible workflow artifacts in the agentic folder tasks.
- Generated a valid standalone City Plan SVG.
Where it failed
- Zero strict resolved practical cases.
- Repeated proof-code and proof.txt failures.
- Evidence and attachment details did not match the hidden oracle.
- One workflow case missed core due normalized text, proof, and warning-code issues.