GLM 5.3 NVFP4 API
Model source: Hugging FaceAn all-rounder for writing and coding, with tool use.
GLM 5.3 is the largest model in our catalogue: a 753-billion-parameter mixture of experts from Z.ai with a one-million-token context, for the writing, coding and agent tasks that smaller models get wrong. On GPU Flow it costs €1.206 per million input tokens and €3.789 per million output, with cached input at €0.224. A default key accepts prompts of up to 256,000 tokens; above that, request it.
Reads in one go
≈ 2,200 pages
1,048,576 tokens; above 256,000 per request, ask us first
Brain size
The largest in the catalogue
753B MoE, 40B active, NVFP4
Understands
Text, not images
Chat, Coding, Reasoning, Tools
What it is good at
- Hard reasoning in several steps. Works through problems that need a plan and several steps before the answer, where smaller models lose the thread.
- Large codebases. Understands them whole and proposes complete changes, not isolated snippets.
- Tool use as the basis of an agent. Calls tools reliably, so it can search, read and act as part of an agent loop.
- Enormous documents. Reads up to 1,048,576 tokens (about 2,200 pages) in one go without losing the thread. A default key cuts the prompt at 256,000 tokens; above that, request it.
Mixture of experts: performs like a huge model without costing like one. If you only need simple chat, classification or translation, Qwen3.5 9B does it for a fraction of the price.
Pricing
€ per million tokens, VAT included
For your technical team
Compatible with the OpenAI SDK: change the base URL and the key, and that is it.
curl https://api.gpuflow.ai/v1/chat/completions \
-H "Authorization: Bearer $GPUFLOW_API_KEY" \
-d '{ "model": "glm-5.3-nvfp4",
"messages": [{ "role": "user", "content": "Hola" }] }'Context and limits
- Context
- 1,048,576 tokens, about 2,200 pages. A default key accepts up to 256,000 prompt tokens; above that, you have to ask.
- Requests per minute
- 1,000 per key, with bursts up to double; raisable to 5,000 from the dashboard. 5,000 in total per account.
- Tokens per minute
- 2,000,000 input (cached tokens do not count) and 400,000 output, per key.
- Prompt per request
- Up to 256,000 tokens, raisable on request.
- If you go over
- A 429 error with a Retry-After header telling you how many seconds to wait.
GLM 5.3 NVFP4 compared with similar models
| GLM 5.3 NVFP4 | Qwen3.8 27B | Nemotron 3.5 Lightning 30B | |
|---|---|---|---|
| Best for | Hard reasoning, large codebases, enormous context | Coding, agents, structured output | Fast reasoning and tool use at low cost |
| Context | ≈ 2,200 pages | ≈ 210 pages | ≈ 280 pages |
| Input / output, € per million | 1.206 / 3.789 | 0.10 / 0.30 | 0.12 / 0.30 |
| 1,000 chat answers | ~1.74 € | ~0.14 € | ~0.15 € |
When the task is hard or the context is over 100k tokens, GLM 5.3; when it is everyday coding, Qwen3.8 27B for a fraction of the price; when it is fast reasoning at scale, Nemotron 3.5 Lightning.
Frequently asked questions
Is GLM 5.3 NVFP4 free?
The model weights are open source; running them costs GPU time. On GPU Flow you pay €1.206 per million input tokens, and every new account gets €5 of credit when it adds a card.
What does the GLM 5.3 NVFP4 API cost?
€1.206 per million input tokens, €3.789 per million output and €0.224 per million cached input. VAT included, no monthly fee.
Does GLM 5.3 NVFP4 support function calling?
Yes, with the standard OpenAI tools parameter.
Can I use GLM 5.3 NVFP4 with Claude Code or Cursor?
Yes. Claude Code connects through the Anthropic-compatible endpoint (/v1/messages); Cursor and other tools through the OpenAI-compatible one (/v1/chat/completions).
How do I use the 1M context?
The model accepts up to 1,048,576 tokens, but a default key cuts the prompt at 256,000 tokens. If you need more, ask us to raise it for your key.
What does NVFP4 mean?
NVFP4 is NVIDIA's 4-bit floating-point format, native to its Blackwell GPUs. We serve an NVFP4 quantisation of GLM 5.3 rather than the original 16-bit weights, which is what makes a model this size affordable to run.