One API keyfor the leading LLMs
Chat, video, audio, embeddings and rerank. Pay as you go, with public pricing.
Why teams build on ModelSite
Direct from vendors
Capacity bought in bulk across major vendor channels and scheduled centrally, so prices come in below going direct. Every model's price is public in the catalog.
- -45%
- at best, off official list price
- $0.00255
- cache-read price from (per 1M tokens)
- qwen3-vl-plus-45%
$0.149$0.082$1.488$0.819$0.017 - glm-5.3-40%
$1.191$0.715$4.167$2.50$0.179 - glm-5.3-flash-40%
$0.12$0.072$0.417$0.25$0.021 - deepseek-v4.1-flash-1017-15%
$0.15$0.128$0.60$0.51$0.00255
GLM-5.3 costs half as much overnight (00:00–08:00 Beijing time): input $0.358 / output $1.25 / cache read $0.09.
DeepSeek V4.1 Flash costs about 2x at weekday peak (09:00–12:00 and 14:00–18:00 Beijing time): input $0.255 / output $1.02 / cache read $0.0051.
- Metering dimensions per call (illustrative)
- Input
- Output
- Thinking
- Cache read
- Cache write
Transparent billing
Public list prices, pay as you go. Break spend down by API key, member or model. Set a monthly budget and requests stop when it runs out.
- No minimum top-up
- Balance never expires
- Clear billing, benchmarked to official list prices
- The only line that changes
client = OpenAI(- base_url="https://api.openai.com/v1"+ base_url="https://api.modelsite.ai/v1")Dual-protocol compatible
Native Chat Completions, Responses and Anthropic Messages endpoints, with streaming intact. Point your existing SDK at a new base_url; no rewrites.
- Stored
- Model name
- Token usage
- Cost
Never stored- Prompts
- Responses
Data security
Prompts and responses are forwarded, never stored. Only the usage summary needed for billing is kept, and traffic is encrypted in transit.
One ingress. Lossless streaming.
Any OpenAI-compatible client enters through a single gateway; the routing core forwards losslessly to upstream clusters.
Curated frontier showcase
Change one request parameter, model: "..." for an instant switch.
Tongyi Qianwen's Qwen3 vision model — thinking and non-thinking modes in one, with image and long-video understanding, visual coding and spatial perception.
Zhipu's flagship text model — 1M context, 128K max output, much stronger at complex software engineering, terminal use and agent tasks, with deep thinking always on.
The Flash edition of Zhipu's GLM-5.3 series — low unit price, built for high-volume calls.
DeepSeek's newest lightweight flagship — 1M context, fast and low-cost, built for high-frequency calls and agents.
Estimate your monthly enterprise savings
Prices come live from the catalog; assumes a 1:3 input/output ratio.
Ready to unify your model access?
Swap one line of base_url and call the leading models with a single key. Transparent pricing, pay as you go, no minimum top-up, balances never expire.