Every AI tool gets tested before I’ll recommend it.
I run the same hard prompts through cloud models, local models and the apps that manage them — then publish what it actually cost. No reviews without receipts.
Latest reviews
DeepSeek V4 Pro
The paid one I reach for when a job spans more than one file.
Worth paying for on the jobs that go live. Not worth using for a one-line fix, where the local model is free and fast enough.
DeepSeek V4.1 Flash
Cheap, fast, and good enough for most of the work I used to think needed a big model.
My default for everyday work. Switch up to V4 Pro the moment a task crosses a file boundary or goes live.
Qwen 3.8 27B (24K, no thinking)
The free one. Fits entirely on a 16 GB card at 24K context with thinking turned off.
If the task fits in one file, start here. It costs nothing and it is fast. Move up only when it visibly struggles.
Qwen 3.8 27B (32K, thinking on)
Same model, more context, thinking switched on. Slower, and the extra room costs you speed.
Keep it installed for hard questions you cannot send off the machine. For everything else the 24K tag is the better trade.
Guides
Make a 27B model fit on a 16 GB card
The exact setup I use for a 27B model at 24K context on an RTX 4080 SUPER, including how to check it really is on the GPU.
Updated 2026-11-08
How I test a model before I recommend it
The five tests, how the score out of 5 is decided, and what the cost figure actually measures.
Updated 2026-11-08
One email a week
What I tested this week, what it cost to run, and the one thing worth your time. No spam, unsubscribe in one click.