GLM 5.3 (sometimes called "GLM 5" or "GLM 5.2", though those are now earlier versions) came out on 14 August 2026 and became Z.ai's main model for code and agent tasks. Its context is 1 million tokens and its output limit is 128 thousand. Next to it runs a lighter GLM 5.3 Flash, and for steady work in the terminal Z.ai sells the Coding Plan with a fixed monthly quota. What follows: how the models differ, what access costs, and what to do if you need code through an API that accepts payment from Russia.
What GLM 5.3 is
Z.ai released GLM 5.3 as an upgrade to GLM 5.2 with no new pretrained model: only the post-training changed, on the same base architecture. Z.ai's documentation confirms a context window of 1M tokens and an output ceiling of 128 thousand. The company does not disclose the parameter count.
Z.ai claims a 50% gain over GLM 5.2 on its own Z.ai Code Bench and leading results on Terminal Bench 3.0. It also stresses cybersecurity: the model found more than 2400 real vulnerabilities in 269 open-source projects and came out almost twice ahead of GLM 5.2 on the ExploitBench test. The takeaway is simple. GLM 5.3 is aimed first of all at agent code and security analysis, not at general chat.
Reasoning is always on, with three effort levels: low, high and max. For hard code tasks the developer recommends max, but then both the response time and the cost of a request go up.
How GLM 5.3 Flash differs
GLM 5.3 Flash came out around the same time, at the end of August, with open weights. It is a hybrid model with 320 billion parameters, of which 18 billion are active for each token. The context is the same as the flagship's, a million tokens. Per token Flash is several times cheaper than the main GLM 5.3, so it gets picked for high-volume work: autocomplete, short agent steps, processing large amounts of text where top quality is not essential.
If a request names an older version (GLM 5.2, 5.1 or GLM 4.7), the Z.ai platform itself reroutes it to GLM 5.3 or 5.3 Flash. You do not need to keep the old models in production: the name in the request can stay as it was, and the current version will answer.
Coding Plan: Lite, Pro and Max tiers
For regular agent work Z.ai offers the Coding Plan subscription instead of per-token billing. The official devpack page gives the starting price as follows: from 18 dollars a month for Lite. Pro and Max are built for more volume and more frequent requests, and cost up to 80 and 168 dollars a month respectively, before the annual-payment discount.
The quota works in two layers. Credits are spent and restored every 5 hours, and on top of that sits a weekly limit that refreshes every 7 days after the subscription starts. All tiers include both GLM 5.3 and GLM 5.3 Flash, plus built-in MCP tools for web search and reading repositories. The Coding Plan works with Claude Code, Cline, OpenCode and several other agent tools through a compatible protocol.
GLM 5.3 and GLM 5.3 Flash API prices
Besides the subscription, Z.ai has ordinary pay-per-token billing. According to the official pricing page, API access costs:
| Model | Input, per 1M tokens | Output, per 1M tokens |
|---|---|---|
| GLM 5.3 | $1.40 | $4.40 |
| GLM 5.3 Flash | $0.15 | $0.50 |
The prices are given as of writing, September 2026, from Z.ai's official documentation. Cached input is cheaper than regular input, and the platform stores the cache free for a limited time. Check the exact terms on the pricing page before you work out a budget.
Is GLM available from Russia
Z.ai has made no direct statement about access for users in Russia, neither on the company blog nor in the documentation. According to users in Russian-language tech blogs, the web chat chat.z.ai opens without a VPN, but paying for the API with a Russian card runs into trouble, and those sources suggest solving it through intermediaries. There is no official confirmation of this, and gray payment schemes are risky: the account can be blocked without warning.
Agent coding: GLM Coding Plan or Claude through codexkey
codexkey has GLM: the catalog holds glm-5.3 ($0.2 / $0.6 per 1M tokens, input / output), glm-5.3-flash ($0.1 / $0.32) and glm-5.2 ($0.12 / $0.32), the price list of 29 October 2026. GLM Coding Plan, Z.ai's subscription with a fixed quota, remains a separate product and is not sold through codexkey. The approach to payment is different: instead of a monthly quota you pay for the tokens actually used, with a Russian card or through SBP (the Russian Faster Payments System), and without a VPN.
Claude Code is set up through environment variables in a couple of minutes:
export ANTHROPIC_BASE_URL="https://api.codexkey.online"
export ANTHROPIC_AUTH_TOKEN="sk_cdx_your_key"
export ANTHROPIC_DEFAULT_OPUS_MODEL="claude-opus-5-5[1m]"
export ANTHROPIC_DEFAULT_SONNET_MODEL="claude-sonnet-5[1m]"
export ANTHROPIC_DEFAULT_HAIKU_MODEL="claude-sonnet-5"
The ANTHROPIC_DEFAULT_* lines are required: without them Claude Code requests its own default models and answers "There's an issue with the selected model". After you start claude, pick a GLM model with the command /model glm-5.3. For comparison: Claude Opus 5 (claude-opus-5) costs $1.08 / $5.4, Sonnet 5 (claude-sonnet-5) $0.52 / $2.6. Detailed setup for Windows, macOS and Linux is in the Claude Code guide. Top-ups start at 1 dollar, with no fees for a legal entity or for closing documents.
If you are choosing between models by price and open weights rather than between GLM subscriptions, we have a separate breakdown, Claude vs Qwen. For coding agent options, including Qwen Code, see the article on Qwen 3.8.
FAQ
When did GLM 5.3 come out?
On 14 August 2026, according to the official Z.ai blog. GLM 5.3 Flash appeared around the same time, closer to the end of August.
How is GLM 5.3 different from GLM 5.2?
Not in architecture: both models share one base. The whole difference in quality comes from an extra post-training phase aimed at code, agent scenarios and cybersecurity.
How much does the GLM Coding Plan cost?
Officially only the starting price is known: from 18 dollars a month for the Lite tier. Pro and Max cost more and give more requests in the 5-hour window and per week. Check the exact tier grid on the subscription page before paying.
Does GLM work from Russia without a VPN?
There is no clear answer. The web chat, according to users, opens without a VPN, but when paying for the API with a Russian card people run into difficulties. Z.ai has not published an official position on this.
Is GLM 5.3 Flash an open model?
Yes, the GLM 5.3 Flash weights are published openly: you can download the model and deploy it yourself if you have enough hardware. GLM 5.3 itself stays closed and is available only through the Z.ai API or chat.
Is GLM available in codexkey?
Yes. The codexkey catalog has glm-5.3, glm-5.3-flash and glm-5.2. If you were looking for GLM for agent code with payment from Russia and no VPN, Claude Code through codexkey fits: a GLM or Claude model is picked with the /model command.