Policy
180 Million Wrong Answers a Month: Why a German Court Just Held Google Responsible
By Daniel R. Whitfield · 8 min
The Alphabet subsidiary is betting 'active inference' can more accurately reconstruct what human drivers would do in crash scenarios—and wants academics to validate it.
| # | Model | Reason | Code | $/M |
|---|---|---|---|---|
| 1 | GPT-5.1 OpenAI | 92 | 88 | 9.0 |
| 2 | Claude Sonnet 4.5 Anthropic | 91 | 90 | 8.4 |
| 3 | Gemini 2.5 Pro Google | 88 | 84 | 6.5 |
| 4 | ERNIE-X 350B Baidu | 84 | 80 | 2.1 |
We pulled filings, planning permits, and grid-connection requests from seven jurisdictions. Here is what the spreadsheet actually says.