GLM-5.2 costs $1.40 per million input tokens and $4.40 per million output tokens on Z.ai's official API, with cached input at $0.26. That rate has not moved since the model launched in June 2026. Everything around it has. As of November 2, 2026, 24 third-party providers list GLM-5.2 on OpenRouter across 33 endpoints, from $0.006 per million input tokens at the bottom to $2.31 at the top, with an official-parity cluster of eleven hosts at exactly $1.40 in the middle.
This is the third shape this market has taken. At launch, hosts clustered near the official rate. By August the spread had opened to 30x, then snapped shut to 3.5x within two weeks when the discount tier repriced. It has since reopened wider than before, this time with a new kind of host at the bottom and a new paid tier at the top. This page has the current official rates, every host grouped by what they charge, what the fast endpoints are, and what a real workload pays on each route. Z.ai's current flagship is GLM-5.3 at the same official price; its small model is covered in GLM-5.3 Flash API pricing.
GLM-5.2 official API pricing
Rates from Z.ai's pricing page, USD per 1M tokens:
| Token type | Z.ai official |
|---|---|
| Input | $1.40 |
| Cached input | $0.26 |
| Output | $4.40 |
The full 1M-token context window bills at the standard rate with no long-context surcharge, which remains the model's standout economic feature for long-document and agent work. Z.ai lists GLM-5.3 at the same $1.40 and $4.40, so on the official API the newer model costs nothing extra. Cached input storage is marked free for a limited time on Z.ai's page.
GLM-5.2 pricing by host
OpenRouter standard routing as of November 2, 2026, grouped by tier. USD per 1M tokens. Promotional discounts are shown at the discounted rate with the discount noted.
| Tier | Hosts | Input | Output | Cached input |
|---|---|---|---|---|
| Near-zero input | Wafer (US), Relace | $0.006 to $0.0065 | $6.80 to $7.91 | $0.005 to $0.0065 |
| Lowest balanced | Morph | $0.078 | $1.40 | $0.069 |
| Low input, high output | Wafer, Cloudflare | $0.19 | $8.00 | $0.18 to $0.19 |
| Promotional | DeepInfra (25% off), StreamLake (56% off), Novita (54% off), SiliconFlow (50% off), AtlasCloud (33% off) | $0.45 to $0.94 | $1.44 to $2.95 | $0.08 to $0.17 |
| Mid market | Ambient, CoreWeave, DigitalOcean | $0.60 to $0.70 | $2.00 to $2.42 | $0.11 to $0.15 |
| Alibaba | Alibaba (two regions) | $0.97 to $1.12 | $3.04 to $3.52 | $0.19 to $0.22 |
| Official parity | Z.ai, Together, Baseten, Nebius, Mistral, Parasail, GMICloud, Venice, Friendli, Baidu, Inceptron | $1.40 | $4.40 | $0.14 to $0.26 |
| EU region | Mistral (EU) | $1.54 | $4.84 | $0.154 |
| Fast tier | Baseten (fast), Baidu (fast), Alibaba (fast) | $2.10 to $2.31 | $6.60 to $7.88 | $0.21 to $0.56 |
Four things to read off the table.
The bottom of the market is not what it looks like. Wafer's US endpoint and Relace price input at well under a cent, but output at $6.80 and $7.91, 1.5 to 1.8 times the official output rate. That structure pays off only for traffic that is almost all input, such as classification or extraction over long documents. For a normal assistant mix it is not the cheapest route. Both run 4-bit quantized weights.
The real floor for a balanced workload is Morph at $0.078 input and $1.40 output, roughly 18 times below official on input and 3 times below on output, on 8-bit weights. One host, so treat it as a price to test, not a price to budget on.
Five hosts are on promotion, from 25 to 56 percent off. StreamLake and Novita are the two that ran the $0.07 tier in August and repriced 10x overnight; both are discounting again. A promotional price is a price with an end date, and a buyer who depends on one should lock capacity before it resets.
The fast tier is new since this page was last updated. Baseten, Baidu and Alibaba each run a higher-throughput endpoint at roughly 1.5 times the official rate, $2.10 to $2.31 input and $6.60 to $7.88 output. These are what searches for "GLM-5.2 fast pricing" are asking about: the same model, same weights, priced for speed. Z.ai itself sells speed on GLM-5.3 the same way, through GLM-5.3 Prime at $2.80 and $8.80 on Alibaba Cloud.
What a real workload pays
A production assistant doing 10M input and 1M output tokens per day, uncached, at list rates on November 2, 2026:
| Route | Daily | Annual |
|---|---|---|
| Morph | $2.18 | $796 |
| Wafer (US), near-zero input | $6.86 | $2,504 |
| DeepInfra (25% promotion) | $7.43 | $2,710 |
| DigitalOcean | $9.20 | $3,358 |
| Z.ai official, or any parity host | $18.40 | $6,716 |
| Baseten fast | $27.60 | $10,074 |
The table makes the near-zero-input point concrete. Wafer's $0.006 input rate runs this workload at $6.86 a day because the $6.80 output rate dominates once output is a tenth of input. Morph, at $2.18, is a third of that and an eighth of the official rate. Flip the mix to 50M input and 100K output, a document-processing pattern, and Wafer drops to about $0.98 a day while Morph is $4.04. Which host is cheapest depends entirely on the shape of your traffic.
Cache does less here than on DeepSeek or MiniMax. At a 60 percent cache hit rate the official route falls from $18.40 to $11.56 a day and Morph from $2.18 to $2.13, because Morph's cached rate is almost the same as its uncached rate. The gap between the floor and the official API stays above 5x at every cache level.
How GLM-5.2 compares
Official rates, USD per 1M tokens:
| Model | Input | Output | Source |
|---|---|---|---|
| GLM-5.2 | $1.40 | $4.40 | Z.ai |
| GLM-5.3 | $1.40 | $4.40 | Z.ai |
| GLM-5.3 Flash | $0.15 | $0.50 | Z.ai |
| GLM-5.3 Prime (high speed) | $2.80 | $8.80 | Alibaba Cloud |
| Mistral Large 4 | $0.70 | $2.10 | Mistral |
| DeepSeek V4 Pro, peak | $1.32 | $3.96 | DeepSeek |
GLM-5.2 and GLM-5.3 share a rate card, so a buyer on the official API has no price reason to stay on 5.2. The third-party market tells a similar story: GLM-5.3 has 48 endpoints on OpenRouter against 33 for 5.2, and its floor is in the same range. At $1.40 and $4.40 the GLM flagships sit above Mistral Large 4 since its October 29 reset and just above DeepSeek V4 Pro at peak. For the model-by-model view see the LLM API pricing comparison.
How to choose a GLM-5.2 host
Balanced assistant traffic: Morph at $0.078 and $1.40 is the floor. Verify precision (8-bit), the full 1M context, and rate limits, and keep a tested fallback.
Input-heavy pipelines: Wafer's US endpoint or Relace. Under a cent on input makes long-document classification nearly free; the $6.80 to $7.91 output rate is the cost of that structure.
Stability over price: the eleven parity hosts at $1.40 have held that rate since June through two spread cycles. Together, Baseten, Nebius and Z.ai itself are the usual picks.
Latency-sensitive: the fast tier at Baseten, Baidu or Alibaba, at about 1.5 times official. Pay it only where tokens per second has a measurable value.
On a promotional host: budget at the undiscounted rate and lock what you can. Two of the five promotional hosts have already repriced 10x once this year.
Budget certainty: a forward contract remains the only way to fix a price in a market that has run from 30x to 3.5x to several hundred times in one quarter. How to lock in token prices covers the mechanics, and why token prices differ covers why a spread this wide exists at all.
To track GLM rates across hosts as they move, see the Mercatus Token Index.
Frequently asked questions
How much does GLM-5.2 cost per million tokens?
$1.40 input and $4.40 output on Z.ai's official API, with cached input at $0.26. Third-party hosts range from $0.006 to $2.31 input and $1.40 to $8.00 output across 24 providers on OpenRouter as of November 2, 2026.
Who is the cheapest GLM-5.2 provider?
It depends on your input-to-output mix. Morph is cheapest for a typical assistant workload at $0.078 input and $1.40 output. Wafer's US endpoint and Relace are cheapest for input-heavy traffic, at under a cent per million input tokens but $6.80 to $7.91 output.
What is GLM-5.2 fast pricing?
Higher-throughput endpoints for the same model at Baseten, Baidu and Alibaba, priced at $2.10 to $2.31 input and $6.60 to $7.88 output, about 1.5 times the official rate.
Why do GLM-5.2 prices differ so much between hosts?
Hosts run the same open weights on different hardware and precision (4-bit, 8-bit), at different utilization and margin, and some are running promotions. Several also shift cost between input and output to suit a traffic pattern, which is why a host can be cheapest on one column and most expensive on another.
Is GLM-5.3 more expensive than GLM-5.2?
Not on the official API. Z.ai lists both at $1.40 input and $4.40 output. GLM-5.3 Prime, the high-speed variant, is $2.80 and $8.80 on Alibaba Cloud.
Does GLM-5.2 charge more for long context?
No. The full 1M-token window bills at the standard rate on the official API, with no long-context surcharge.
Methodology
Official GLM-5.2, GLM-5.3 and GLM-5.3 Flash rates from Z.ai's pricing page, read November 2, 2026. Third-party host prices, cached rates, precision, and promotional discounts from OpenRouter's GLM-5.2 endpoint data, standard routing, November 2, 2026; 33 endpoints from 24 providers, grouped into tiers above, with ranges spanning the hosts named in each row. GLM-5.3 endpoint count and GLM-5.3 Prime rate from OpenRouter the same day. Earlier market states (30x spread in early August, 3.5x in late August) from previous versions of this page. Mistral and DeepSeek rates from the linked Mercatus pages. Workload examples use stated token volumes at 365 days and ignore cache write charges; recompute with your own mix. Last verified: 2026-11-02.