Quick Answer
Two free tiers — deepseek-v4-pro-free and deepseek-v4-flash-free — each give 200 requests/day, per-account and reset daily. After running both, the only sensible split is:
deepseek-v4-pro-free(200/day): multi-file reasoning, long agent loops, anything where a wrong answer costs you a redo.deepseek-v4-flash-free(200/day): the volume work — explain-a-function, commit messages, single-file edits, quick lookups.
One line: route by difficulty, not habit. Here's the breakdown.
What Goes to Pro
Pro's value is getting it right the first time. Anything where a wrong answer means redoing, rerunning, or another round goes to Pro:
- Multi-file reasoning — cross-module changes that require understanding the repo structure.
- Long agent loops — build + self-test + fix, where reasoning depth drives success rate.
- Hard problems — hard-to-reproduce bugs, deep architecture questions.
200 real Pro turns comfortably cover a normal day's hard problems — hard work is few and focused, not won by volume.
What Goes to Flash
Flash's value is cheap + fast, for high-volume deterministic work:
- Explain a function / an error;
- Write commit messages, update docs;
- Single-file edits;
- Quick lookups ("how does this syntax work", "this library's API signature").
200/day absorbs basically all of this high-frequency light work. Handing it to Flash preserves your Pro quota.
The Classic Mistake: Burning Pro on Trivia
Most people default to Pro and drain the 200 pro-free requests before lunch — explanations, small edits, syntax lookups — then hit empty exactly when they need it for real problems.
Routing by difficulty isn't mystical; it's a habit you can execute: default light work to Flash, and switch to Pro only when genuinely stuck or when the cost of a wrong answer is high. Together, the two free tiers cover nearly a full day for free.
The Fallback When Quota Runs Out
Once either free tier hits zero, the matching paid model has no account-level limit:
| Free tier | Paid fallback | Price (per 1M tokens) |
|---|---|---|
deepseek-v4-pro-free |
deepseek-v4-pro |
$1.74 / $3.48 flat |
deepseek-v4-flash-free |
deepseek-v4-flash |
$0.14 / $0.28 flat |
Paid Flash stays at the old rate, and Pro is flat with no peak windows — so exceeding the free quota isn't painful; it just goes from zero to a few dollars.
A Routing Rule You Can Actually Use
if task == light (explain / small edit / commit / lookup):
use deepseek-v4-flash-free
elif task == hard (multi-file / long loop / high-stakes):
use deepseek-v4-pro-free
elif either quota == 0:
use the matching paid model (deepseek-v4-pro / deepseek-v4-flash)
This rule holds in any multi-model tool (dsh, Codex, Cherry Studio, etc.): configure both free tiers, then set your default routing.
FAQ
How different are Pro and Flash?
Flash is a 284B MoE with 13B active params, built for "fast + cheap + high-frequency"; Pro targets the capability ceiling. Daily high-frequency on Flash, hard problems on Pro is the standard combo.
Can I use both free tiers at once?
Yes. They're two independent 200/day pools that don't consume each other — which is the whole premise of difficulty routing.
Does the quota reset at midnight?
Documented as daily reset; check the rate-limits docs for specifics.
I'm unsure if a task is light or hard — what do I do?
Run it on Flash first; if quality falls short or a wrong answer would force a redo, switch to Pro. It's not waste — Flash is cheap, so trial cost is near zero.
Can routing be automated?
Tools that support it can route by rule (e.g. gateway-layer multi-model routing); manual "light on Flash, hard on Pro" is enough for solo developers.
Register at TeamoRouter — two free tiers, 200/day each, routed by difficulty, nearly free V4 all day.