# GLM 4.5 Air

> Cheap agentic MoE (106B/12B active).

Human version: https://kymaapi.com/models/glm-4.5-air
Live JSON: `GET https://kymaapi.com/v1/models` (no auth required)

## Facts

- **Model ID**: `glm-4.5-air` — pass this as `model` in the request body
- **Creator**: Zhipu AI
- **Released**: 2025-07-28
- **Context window**: 131K tokens
- **Max output**: 8K tokens per response — a hard ceiling, not a default
- **Price**: $0.176 in / $1.148 out per 1M, cached input 10%
- **Capabilities**: tools, reasoning, caching

## Call it

```bash
curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "glm-4.5-air", "messages": [{"role": "user", "content": "Hello"}]}'
```

## Positioning

GLM 4.5 Air is a fast, low-cost mixture-of-experts model built for high-throughput agent workflows and long-context text processing. Reach for it when you need reliable tool calling and reasoning at scale without premium token costs.

## About

Created by Zhipu, GLM 4.5 Air is a mixture-of-experts architecture with 106B total parameters and 12B active parameters per forward pass. It handles text-only inputs and outputs, supports tool calling, native reasoning, and structured JSON outputs, and operates in the fast speed tier.

On Kyma, the model accepts up to 131,072 context tokens and generates a maximum of 8,192 output tokens. Every request benefits from automatic failover, and responses include exact token costs in the usage.cost field alongside an X-Kyma-Model header confirming the active routing path.

Prompt caching is fully supported, billing repeated system prefixes at 10% of the standard input rate. The model does not support vision or multimodal inputs, and the 8,192 output limit restricts it from generating extremely long documents in a single turn.

## Use cases

- **High-Volume Agent Orchestration** — Run concurrent tool-calling loops where low latency and cheap tokens matter most.
- **Long Context Text Analysis** — Process and summarize large documents or conversation histories within a 131k token window.
- **Structured Data Extraction** — Parse unstructured text into strict JSON formats using built-in structured output support.
- **Fast Conversational Routing** — Handle high-throughput chat workloads that prioritize speed and cost over peak reasoning depth.

## Not ideal for

Do not use this model for tasks requiring image understanding, multimodal inputs, or single-turn outputs exceeding 8,192 tokens.

## Pick something else when

- You need image or video understanding → `gemini-3.5-flash`, `gpt-5.6-terra`
- You need outputs longer than 8,192 tokens → `qwen-3.7-max`, `deepseek-v4-pro`
- You need maximum reasoning accuracy over speed → `deepseek-r1`, `qwen-3.7-plus`

## See also

- All models: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md
- Other models by Zhipu AI: https://kymaapi.com/models?q=Zhipu%20AI