GLM 5.3 Flash versus DeepSeek Flash: cheaper by the hour, and $10 free

We have opened a free tier on above.dev: $10 of API credit for GLM 5.3 Flash, no card, for the first 100 accounts. This post explains what the model is, why we picked it for the offer, and how its pricing compares with DeepSeek V4 Flash, the model most people reach for when they want a cheap agentic workhorse.

What GLM 5.3 Flash is

GLM 5.3 Flash is Z.ai's small reasoning model from the GLM 5 family. It has a 1M token context window, native tool calling, structured output, and vision, and it thinks before it answers. On coding-agent traffic it behaves like a model a tier above its price: it follows multi-step plans, reads large diffs without losing the thread, and its tool calls are well formed. It is not a frontier model, and it will not out-reason DeepSeek V4 Pro or Kimi K3. It is the model you run for the 90% of agent turns that are routine.

The price, next to DeepSeek V4 Flash

All figures are what above.dev charges per million tokens. DeepSeek V4 Flash has two rates depending on the time of day in Beijing, so it gets two rows.

Model	Input	Cache hit	Output
GLM 5.3 Flash	$0.165	$0.032	$0.55
DeepSeek V4 Flash, off-peak	$0.242	$0.0077	$0.726
DeepSeek V4 Flash, peak	$0.484	$0.0154	$1.452

On input and output, GLM 5.3 Flash is cheaper than DeepSeek Flash at any hour, by about a third off-peak and by nearly two thirds during DeepSeek's peak window. DeepSeek wins on one number: its cache-hit rate is about four times lower. Whether that matters depends on your cache-hit ratio, so here is a worked session.

A worked agent session

Take a typical coding-agent hour: one million input tokens, 20,000 output tokens, with 90% of the input served from the prompt cache. That cache ratio is what we see from real agent traffic on the gateway.

Model	Session cost
GLM 5.3 Flash	$0.056
DeepSeek V4 Flash, off-peak	$0.046
DeepSeek V4 Flash, peak	$0.091

So the honest summary is this. If you can schedule your work around Beijing business hours, DeepSeek Flash off-peak edges it by a cent an hour. If you cannot, or you do not want to think about it, GLM 5.3 Flash is cheaper than DeepSeek Flash whenever the peak rate applies, and it costs the same all day. It also has vision and a 1M context, which DeepSeek Flash does not.

Why the free tier runs on this model

We serve GLM 5.3 Flash from two routes. Paying customers use Fireworks. The free tier uses a dedicated endpoint we host on Modal, priced identically, so the credit buys exactly the volume a paying customer gets. Both routes pass the upstream cache discount straight through, and our testers on the free route are seeing cache-hit rates above 90%. At those rates, $10 covers roughly two weeks of daily coding-agent use.

How to claim it

Go to above.dev/free, enter your email, and click the link we send. Your key arrives by email and appears in your dashboard. Point any OpenAI-compatible client at https://api.above.dev/v1 with model glm-5.3-flash-modal. When you buy a credit pack later, every model in the catalog unlocks on the same key.

One grant per person, new accounts only, and the first 100 accounts. The counter on the page is live.