I am surprised by how good Gemini flash 05-20 is it seems to be great at solving bugs and general use in agent mode.
The timing couldn’t be better. I’ve found myself needing to conserve my premium fast requests lately, especially after the recent rework of the Slow queue, which now feels truly slow and excludes Sonnet 4.
How’s everyone else’s experience so far?
It’s not bad in its coding skills but sadly breaks a lot with tools usage. It’s still my free model of choice.
Btw, soon there should be a special coding variant of GPT 4.1 model available in the open, and 4.1 is the free base model used in Github Copilot, so I expect them to update it to this new one. If this one is a significant improvement for agentic / tools / MCP use, this might be a big thing. Too bad it’s 1 request in Cursor.