GLM-4.5-Air.
Smaller, cheaper sibling of GLM-4.5. 106B total, 12B active.
1× Nvidia A100 · $0.48/hr.
Most-aggressive quantisation we have a working recommendation for. Lower precision = less VRAM = cheaper hardware, at a small accuracy cost.
Cheapest hosted endpoints.
Speed across providers.
Tokens/sec and time-to-first-token measured against the same prompt template on each provider's API.
| Provider | Tokens/sec | TTFT | Total |
|---|---|---|---|
| OpenRouter | 19.1 | 1992 ms | 4979 ms |
Variants in the GLM family.
GLM-4.5
Zhipu's frontier open-weight MoE — 355B total, 32B active. Strong agentic + r...
GLM-5.1
Zhipu's GLM 5.1 series — successor to GLM-5 on z.ai's API.
GLM-5
Zhipu's GLM 5 generation — closed flagship between GLM-4.7 and GLM-5.1.
GLM-5 Turbo
Faster, cheaper sibling of GLM-5 on z.ai.
GLM-4.7
Mid-generation GLM 4.7 released between GLM-4.6 and GLM-5.
GLM-4.6
Incremental upgrade on GLM-4.5 — improved reasoning, same context window.
GLM-4.5-Flash
Lowest-latency, lowest-cost variant of GLM-4.5 on z.ai.
Frequently asked.
How do I run GLM-4.5-Air?
Where can I access GLM-4.5-Air?
How much does it cost to run GLM-4.5-Air?
Is GLM-4.5-Air open-source or proprietary?
Cheapest hardware per quantisation.
Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/hr refreshed hourly from each provider's API.
| Quantisation | Cheapest GPU config | Total VRAM | Live $/hr | tokens/sec | |
|---|---|---|---|---|---|
|
FP16
FP16 — half precision (default)
|
288 GB | — | — | Compare → | |
|
FP8
FP8 — 8-bit float (Hopper / Blackwell)
|
141 GB | — | — | Compare → | |
|
INT4
INT4 — 4-bit integer (~4× VRAM saving)
|
80 GB | $0.48/hr | — | Compare → |
What it costs per month across providers.
Estimate your monthly bill for GLM-4.5-Air across every host that publishes per-token pricing. Slide your token volumes; the chart + table re-rank cheapest-first.
Cheapest provider on the left.
Total monthly cost — input + output tokens combined.
Bill breakdown.
Rent the GPU instead of paying per token.
For an open-weights model like GLM-4.5-Air, you can rent a GPU and serve inference yourself. The math: cheapest GPU rental × 730 hours/month + your electricity rate × power draw.
Assumes the GPU runs 24/7 at ~85% utilisation. If your traffic is bursty, you'll pay less for the API and probably more for the GPU (idle hours still cost rental). The breakeven analysis lives on the Self-host vs API breakeven tool.
About GLM-4.5-Air.
GLM-4.5-Air is the lightweight variant of Zhipu's GLM-4.5 launch. 106B total parameters with 12B active per token — about a third the size of the flagship, with a corresponding drop in benchmark scores but much cheaper inference. Same architecture, training recipe, and tool-use support as the full model. Targeted at production deployments where the flagship's parameter count would make per-token cost prohibitive. Open-weight under MIT.
How it's built.
How much it can remember.
What it can do.
Every place this model is hosted.
Self-hosted on rented GPU cluster
self hostedSmaller MoE — single-node deployment feasible.
Self-hosted on rented GPU cluster
self hostedSmaller MoE — single-node deployment feasible.