So I run all tasks through four models locally: Astra X-High with Codex, Fable 5.2 Extra with Claude Code, Carbon High with Antigravity, and Grok 4.7 High Fast with Cursor.
I would not put Google in the same league as OpenAI and Anthropic. It is like 3 times slower at completing tasks. It also has the highest failure rate out of any of the models. Like sometimes it can do what Astra and Fable can do but sometimes it just misses completely.
Nah, GPT6 models have the highest failure rate - they simply stop, repeatedly and consistently.
And obviously you don't run all the tasks through these models *locally*