TL;DR — As of August 2026, Zhipu's official pay-as-you-go pricing (open.bigmodel.cn) runs from free (GLM-4.7-Flash, 200K context) to ¥8 in / ¥28 out per million tokens for the flagship GLM-5.2. Pricing is tiered by request input length — short requests bill at a lower tier — and cache hits cost roughly 20–25% of the input rate. In USD terms the flagship works out to roughly $1.40 in / $4.40 out, which undercuts Western frontier models by 3–10× and is why GLM has become a serious routing target for cost-sensitive workloads.
Official GLM pricing table (August 2026)
Rates in CNY per million tokens, from Zhipu's official platform (open.bigmodel.cn), current as of 2026-08-09:
| Model | Input | Output | Cache hit | Notes |
|---|---|---|---|---|
| GLM-5.2 | ¥8 | ¥28 | ¥2 | Flagship |
| GLM-5.1 (short requests) | ¥6 | ¥24 | ¥1.3 | 200K context, 128K max output |
| GLM-5.1 (long requests) | ¥8 | ¥28 | ¥2 | |
| GLM-5-Turbo (short) | ¥5 | ¥22 | ¥1.2 | |
| GLM-5-Turbo (long) | ¥7 | ¥26 | ¥1.8 | |
| GLM-5 (short) | ¥4 | ¥18 | ¥1 | |
| GLM-5 (long) | ¥6 | ¥22 | ¥1.5 | |
| GLM-4.7 | ¥2–4 | ¥8–16 | ¥0.4–0.8 | Tier depends on input length |
| GLM-4.5-Air | ¥0.8–1.2 | ¥2–8 | ¥0.16–0.24 | Lightweight |
| GLM-4.7-FlashX | ¥0.5 | ¥3 | ¥0.1 | Fast tier |
| GLM-4.7-Flash | Free | Free | Free | 200K context |
Three things to know before estimating a bill:
- Tiered billing: the input length of each individual request determines its price tier — a mostly-short-request workload bills meaningfully cheaper than the headline rate.
- Cache pricing is aggressive: cache hits bill at ~20–25% of the input rate, so stable-prefix prompts (agent system prompts, RAG preambles) get a large effective discount.
- Prices moved up in 2026: GLM-5's launch (February 2026) came with a reported 67–100% API price increase over the previous generation — worth remembering if you're comparing against older blog posts.
USD equivalents and how GLM compares
At current exchange rates the flagship lands around $1.40 input / $4.40 output per million tokens. Against the current market (full pricing guide):
| Model | Input /1M (USD) | Output /1M (USD) |
|---|---|---|
| Claude Sonnet 4 | $3.00 | $15.00 |
| GPT-5.4 | $2.50 | $15.00 |
| GLM-5.1 | ~$1.40 | ~$4.40 |
| Gemini 3 Pro | $1.25 | $5.00 |
| DeepSeek V4 Pro | $1.74 | $3.48 |
GLM-5.1 sits in the value tier alongside Gemini 3 Pro and DeepSeek V4 Pro — a fraction of Western frontier pricing, with a genuinely free 200K-context model (GLM-4.7-Flash) at the bottom of the ladder that has no Western equivalent.
Where GLM fits in a routing strategy
For smart routing setups, GLM's ladder maps cleanly to task tiers:
- GLM-4.7-Flash (free): classification, summarization, background/heartbeat traffic — zero-cost capacity
- GLM-5-Turbo / GLM-4.7: everyday drafting and mid-difficulty tasks
- GLM-5.1 / 5.2: reasoning-heavy calls where you still want value pricing
ClawRouters includes GLM models in its routing pool, so requests that fit GLM's strengths route there automatically while quality-critical calls still reach frontier models — you get the price ladder without managing another provider account. See which providers let you mix models to reduce costs.
Routing decision card
The exclusive layer: where GLM models actually land in routing decisions.
- Route here when: value-tier reasoning (GLM-5.1 at ~$1.40/$4.40 competes with Gemini 3 Pro and DeepSeek V4 Pro), or zero-cost background work — GLM-4.7-Flash's free 200K-context tier has no Western equivalent for heartbeat/classification traffic
- Tiered-billing edge: GLM's per-request input-length tiers mean short-request workloads bill below headline rates — routers that keep prompts lean capture this automatically
- Head-to-head: vs DeepSeek V4 Pro ($1.74/$3.48) — GLM wins input price, DeepSeek wins output price; the winner depends on your input/output ratio
- Typical pairing: GLM-4.7-Flash for free bulk → GLM-5.1 or DeepSeek V4 Pro for value reasoning → frontier only on escalation
Pricing sourced from Zhipu's official platform and current Chinese-language pricing summaries as of August 2026. Zhipu adjusts rates with model launches — check open.bigmodel.cn/pricing for the live table before committing to volume.