Skip to content
这个页面有中文版。
切换到中文
API Pricing

Z.ai: GLM 4.5 Air API pricing on OpenRouter

id
z-ai/glm-4.5-air
z-ai
First seen 2026-11-07 13:53 UTC
Open on openrouter.ai/z-ai/glm-4.5-air

RouteThird-partyvia Z.AI + Novita + SiliconFlow

Basis: OpenRouter endpoints: Z.AI (z-ai): official; Novita (novita): third_party; SiliconFlow (siliconflow): third_party Archived copy

Current prices

Per million input tokens1

VariantUSDPlatform’s own wordingSince
via Novita · bf16$0.13/M inprompt: 0.00000013 / token (endpoint novita/bf16)

Per million output tokens1

VariantUSDPlatform’s own wordingSince
via Novita · bf16$0.85/M outcompletion: 0.00000085 / token (endpoint novita/bf16)

Per million cached input tokens (read)1

VariantUSDPlatform’s own wordingSince
via Novita · bf16$0.025/M cache readinput_cache_read: 0.000000025 / token (endpoint novita/bf16)

Change log

WhenModel · what it coversChange
2026-11-09 14:03 UTCPrice changedCache read$0.025/M cache read → $0.03/M cache read+20%input_cache_read: 0.000000025 / token → input_cache_read: 0.00000003 / token
2026-11-09 14:03 UTCPrice changed$0.85/M out → $1.1/M out+29%completion: 0.00000085 / token → completion: 0.0000011 / token
2026-11-09 14:03 UTCPrice changedFlagged$0.13/M in → $0.2/M in+54%prompt: 0.00000013 / token → prompt: 0.0000002 / token

Earlier versions

Model · what it coversUSDPlatform’s own wordingSince
$0.13/M inprompt: 0.00000013 / token2026-11-07 13:53 UTC → 2026-11-09 14:03 UTC
$0.2/M inprompt: 0.0000002 / token2026-11-09 14:03 UTC → 2026-11-11 16:55 UTC
$0.85/M outcompletion: 0.00000085 / token2026-11-07 13:53 UTC → 2026-11-09 14:03 UTC
$1.1/M outcompletion: 0.0000011 / token2026-11-09 14:03 UTC → 2026-11-11 16:55 UTC
Cache read$0.025/M cache readinput_cache_read: 0.000000025 / token2026-11-07 13:53 UTC → 2026-11-09 14:03 UTC
Cache read$0.03/M cache readinput_cache_read: 0.00000003 / token2026-11-09 14:03 UTC → 2026-11-11 16:55 UTC

Where these prices come from

Read from the platform’s public pages and APIs. Every read is archived, so any price here can be checked against what the platform showed at the time.