GLM-5.3 Opens Up to Developers at the Same Price as Its Predecessor
Chinese startup Z.ai's new frontier model is now available via API, matching the old per-token rate while scoring higher on independent benchmarks. But a quirk in how the model writes means your bill may still rise.
Key points
- GLM-5.3, a large language model from Chinese startup Z.ai, became available through the company's API as of this week.
- Z.ai is charging $1.40 per million input tokens and $4.40 per million output tokens, identical to the GLM-5.2 pricing.
- Independent benchmarker Artificial Analysis scored GLM-5.3 at 60 on its Intelligence Index, tying Kimi K3 as the top-performing open-weights model in the world.
- Despite flat token prices, real-world costs per task are estimated to be higher because GLM-5.3 produces longer responses than its predecessor.
- Z.ai says it plans to release the model's weights publicly, but has not confirmed a date or licence terms.
A new AI model that reportedly found a security flaw in the code-editing tool Cursor before it was publicly known is now open for business. GLM-5.3, a large language model (the same technology that powers chatbots like ChatGPT) from Chinese startup Z.ai, launched last week. This week it became callable through Z.ai's API, short for application programming interface, the technical channel that lets developers wire an AI model into their own products.
The price is $1.40 per million input tokens and $4.40 per million output tokens. A token is roughly three-quarters of a word, so a million covers hundreds of thousands of words of conversation. Cached input, text the system has already processed and stored, costs $0.26 per million tokens, and cached-input storage is listed as free for a limited time.
How does the price compare to other models?
GLM-5.3 sits in the mid-range of the current API market, cheaper than the flagship models most people have heard of, but not the cheapest option available.
Using VentureBeat's comparison of one million input tokens plus one million output tokens, GLM-5.3 comes to $5.80. That puts it below Grok 4.6 ($8.00 at its lower context rate), Kimi K3 ($18.00), Claude Opus 5 from Anthropic ($30.00), and GPT-5.6 Sol from OpenAI ($35.00). Google's Gemini 3.7 Flash, at $4.50 through the end of 2026, and OpenAI's GPT-5.6 Luna, at $1.40, both undercut it on headline price.
| Model | Input ($/1M) | Output ($/1M) | Combined |
|---|---|---|---|
| GPT-5.6 Luna (OpenAI) | $0.20 | $1.20 | $1.40 |
| Gemini 3.7 Flash, to Dec 31 2026 (Google) | $0.75 | $3.75 | $4.50 |
| GLM-5.3 (Z.ai) | $1.40 | $4.40 | $5.80 |
| Grok 4.6 under 200K tokens (xAI) | $2.00 | $6.00 | $8.00 |
| Claude Opus 5 (Anthropic) | $5.00 | $25.00 | $30.00 |
| GPT-5.6 Sol (OpenAI) | $5.00 | $30.00 | $35.00 |
Should developers expect the same bill as before?
Not necessarily. The per-token rate is unchanged from GLM-5.2, but Artificial Analysis found GLM-5.3 more verbose: it writes longer answers to the same questions. That pushes the estimated cost per completed task from roughly $0.44 with GLM-5.2 to about $0.68 with GLM-5.3, even though the price list looks identical. We first tracked Artificial Analysis benchmarks on 21 August, in our report on Alibaba's Qwen3.8-27B, and the verbosity gap between generations is a pattern worth watching.
For businesses building tools on top of AI, that gap matters. A customer-service bot or a coding assistant runs thousands of conversations a day, and a 50 percent rise in words per answer can translate directly into a 50 percent higher monthly invoice with no change to the advertised rate.
What happens next?
Z.ai says it intends to release the model's weights publicly, letting anyone download and run GLM-5.3 without paying per token. No date and no licence terms have been confirmed, so developers should treat that as a promise rather than a product. Developers who already subscribed to Z.ai's GLM Coding Plan are currently restricted to the OpenAI Chat Completions-compatible protocol, a standard message format that many tools already support.
Common questions
What does "open weights" mean for ordinary people?
Open weights means the core files of the model are shared publicly, so anyone with enough computing power can run it on their own servers rather than paying Z.ai per use. For most people it changes nothing directly, but it tends to bring prices down across the industry over time.
Is GLM-5.3 actually as good as the expensive models it is being compared to?
On Artificial Analysis's Intelligence Index, GLM-5.3 scores 60, tying Kimi K3 at the top of the open-weights category. That's a strong result, though benchmark scores measure specific tasks and real-world performance varies by use case.
