Hawiyat API
Self-hosted models,
priced per token
GLM 5.3 Flash, DeepSeek V4.1 Flash and GLM 5.2 on one key. Balance topped up from the dashboard, spent per request.
3 models · 1M context · pay per 1M tokens
Uptime
99.9%
confirmed on every line
Cache hit
98.7%
on repeated context
Latency
Low
served from local GPU
Pricing
Half
of official rates
Three models, one key
Prices per 1M tokens, in dollars. Cache rows are billed separately from fresh input.
GLM 5.3 Flash
The workhorse. Cheapest line we run, built for volume.
- Input
- $0.06
- Output
- $0.21
- Cache read
- $0.017
DeepSeek V4.1 Flash
High-throughput line for agents and long pipelines.
- Input
- $0.15
- Output
- $0.60
- Cache hit
- $0.003
GLM 5.2
Flagship line for long-horizon, project-scale work.
- Input
- $0.60
- Output
- $2.10
- Cache read
- $0.15
DeepSeek V4.1 Flash shown at peak rates.
How a request is carried
One key, several lines
OUR GPU
primary
OFFICIAL
fallback 1
OFFICIAL
fallback 2
These models already run on rented GPU capacity that we operate ourselves. That capacity is one line in a fallback chain that also includes the official providers, so a request is always carried by a line that is up. Confirming uptime on several lines at once is what lets us hold availability at 99.9% instead of betting the service on a single endpoint. It is also why you keep our prices: you are not paying the expensive platform rates on top.
Top up from the dashboard
No subscription and no monthly cap to fit inside. You add balance to your key, and every request draws from it at the rates above. Usage, spend and cache savings sit in the same place.
Questions
How is the Hawiyat API priced?
Per million tokens, in dollars, with no monthly minimum. You top up a balance in the dashboard and it is spent on the models you call. Nothing expires and there is no subscription to cancel.
Which models are available?
GLM 5.3 Flash, DeepSeek V4.1 Flash and GLM 5.2 run on our own GPU capacity, all three at half the official rate. All three carry a 1M-token context window and are served from the same endpoint with the same API key.
How do you keep uptime at 99.9%?
The models already run on our own rented GPU capacity, and that capacity is one line in a fallback chain that also includes the official providers. If a line degrades, the request is carried by another one, so the service stays up without us charging platform-level prices.
What does 'optimized to half' mean?
Our caching and request-reuse layer cuts what you pay for the same work. Repeated context is served from cache at a fraction of the input price, so the effective cost of a running workload lands near half.
How do I add balance?
Top up from the usage dashboard. Pick an amount, pay, and the balance is credited to your API key. Usage is metered per request and visible in the same dashboard.