Draft build — noindex ON, scores and costs are placeholders until the test numbers are filled in.

All tools

DeepSeek V4 Pro

The paid one I reach for when a job spans more than one file.

4.5 Cloud model

Tested 2026-11-08

V4 Pro is the model I pay for. It is not the cheapest thing in my stack and it is not the one I use for a quick script. It is the one I switch to when a fix touches six files at once, when a bug only shows up in the gap between two systems, or when the answer has to be right the first time because it is going live.

Pricing is $0.66 per million input tokens and $1.98 per million output tokens, with cached input at $0.022 per million. That cache number is the whole reason it stays affordable for me: a long set of notes read once and reused all session costs almost nothing after the first pass.

Pros

Cons

What I tested

TestResult
Long document summaryTODO: run the 40-page test doc and record time plus any dropped facts.
Coding taskMulti-file refactor across a static site generator: edits landed in all changed files, nothing outside the task touched. First run.
Hard reasoningTODO: record the multi-step question, the answer, and whether it was right.
Image understandingTODO: not tested yet. Text-only prompts so far.
Cost for a week of real usePriced at $0.66/M in, $1.98/M out, $0.022/M cached in. A 100K-token prompt with a 5K-token answer costs about $0.08 before caching.

Bottom line: 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.

What this site is

I test AI tools myself — cloud models, local models, and the apps that manage them. Reviews only appear once I’ve run the same prompts through each tool on my own machine or account. No sponsorships, no affiliate links, no ads. More about how this works.