GLM 5.3 Flash OpenCode is one of the easiest ways to try a strong open coding model right now, because OpenCode offers GLM-5.3-Flash free for a limited time inside its Zen service. GLM-5.3-Flash comes from Z.ai. It is a sparse model with 320B total parameters and 18B active, a 1M-token context window and an MIT licence. In the video below I set it up in OpenCode and put it to work on real tasks. This guide covers the two ways to run it, what it costs, how good it is and where the limits are.
Key takeaways
- OpenCode Zen has a free GLM-5.3-Flash model for a limited time, and it only works inside OpenCode.
- On the Z.ai API it is paid: $0.15 input and $0.50 output per million tokens.
- 1M context, 128K max output, image, video and file input.
- Open weights under MIT, so you can self-host it with vLLM or SGLang.
What is GLM-5.3-Flash?
GLM-5.3-Flash is a model from Z.ai (zai-org on Hugging Face). The model card lists 320B total parameters with 18B active per token, a hybrid of sparse and linear attention, and calls it the first natively multimodal model in the GLM-5 series. It is released under the MIT licence and supports English and Chinese.
Z.ai says it is approaching Claude Opus 4.8 on coding and agentic benchmarks, beats GLM-5.2, and costs about one-tenth of GLM-5.2’s price. Before launch it was tested anonymously as “ox-alpha” on OpenCode and OpenRouter, according to the Z.ai docs. So if you used ox-alpha in OpenCode, you have already tried it.
| GLM-5.3-Flash at a glance | |
|---|---|
| Parameters | 320B total, 18B active |
| Context | 1M tokens, 128K max output |
| Input | Text, image, video and file |
| Thinking | Always on; reasoning effort low, high or max (default max) |
| Features | Tool calling, streaming, context caching, structured output |
| Licence | MIT |
| API ids | glm-5.3-flash, glm-5.3-flashx |
Free vs paid: the two ways to use it in OpenCode
There are two routes, and it helps to know the difference before you start.
Route 1: OpenCode Zen (free for now)
OpenCode’s Zen lists GLM-5.3-Flash Free with the id glm-5.3-flash-free. It is free for a limited time and only works inside OpenCode. You don’t need a Z.ai account. If you don’t want free models to show up, set OPENCODE_DISABLE_FREE_MODELS=1.
Route 2: your own Z.AI key (paid)
If you want it outside OpenCode, or you want it to keep working when the free window ends, use a Z.ai API key. Per the Z.ai pricing page, GLM-5.3-Flash costs $0.15 per million input tokens, $0.03 cached and $0.50 output. It is not free on the Z.ai API. Z.ai’s Coding Plan also gives 3x the quota compared with GLM-5.3, with off-peak usage at 50% of the points.
| OpenCode Zen free | Z.AI API key | |
|---|---|---|
| Cost | Free for a limited time | $0.15 in / $0.50 out per 1M |
| Where it works | Only inside OpenCode | Any OpenAI-compatible tool |
| Account | OpenCode | Z.ai key |
| Best for | Testing and learning | Daily work and other tools |
Want cheap AI coding models working inside your business? Inside the AI Profit Boardroom you get the Agent OS — Hermes, OpenClaw, OpenCode, Claude and Codex in one place — plus step-by-step Hermes training and four live coaching calls a week.
Want a plan for your business first? Claim your free strategy session.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →
How to set up GLM 5.3 Flash in OpenCode
Here is the full setup, taken from the OpenCode docs and the providers page.
- Install OpenCode:
curl -fsSL https://opencode.ai/install | bashornpm install -g opencode-ai. - Open your project and run
opencode. - Free route: run
/connect, choose the OpenCode provider, then/modelsand pick GLM-5.3-Flash Free. Zen models use the formatopencode/<model-id>. - Paid route: get a key at z.ai/manage-apikey/apikey-list, run
/connect, search Z.AI (or Z.AI Coding Plan), paste the key, then/models. - Your key is saved in
~/.local/share/opencode/auth.json.
Next, run /init. OpenCode reads your project and writes an AGENTS.md file, so the model knows how your code is laid out. Press Tab to switch between Plan mode (think first, no edits) and Build mode (make changes). If you don’t like a change, /undo rolls it back and /redo puts it back.
Want an AI coding setup that actually saves you time? Claim your free strategy session →
How good is it? The benchmarks
These are Z.ai’s own numbers from the model card and docs. Treat them as a guide, then test it on your own work.
| Benchmark | GLM-5.3-Flash | Compared with |
|---|---|---|
| Terminal-Bench 2.1 | 84.3 | — |
| DeepSWE v1.1 | 63.4 | GLM-5.2: 46.2 |
| AutomationBench | 48.8 | GLM-5.2: 26.2 |
| Z.ai Code Bench | 29.0 | Opus 4.8: 29.5 |
| ALE CLI pass | 26.3 | — |
| Artificial Analysis Intelligence Index | 57 | — |
The big jump is against GLM-5.2: DeepSWE up from 46.2 to 63.4 and AutomationBench nearly doubled. On Z.ai’s own coding test it lands within half a point of Opus 4.8. In my video it handled real edits and multi-step tasks well for a model this cheap.
Use OpenCode agents with GLM-5.3-Flash
OpenCode has two main agents, Build and Plan, plus subagents called General and Explore. You can call a subagent with @general. You can also make your own agents with opencode agent create, stored in .opencode/agents/ or ~/.config/opencode/agents/, according to the agents docs.
A simple pattern that works: keep a bigger model for planning, and give GLM-5.3-Flash to a custom “builder” agent for the edits and tests. Each agent can have its own provider/model-id, so you only pay top prices where you need them.
Self-host GLM-5.3-Flash
Because the weights are open, you can run it yourself. The model card shows vllm serve "zai-org/GLM-5.3-Flash", and SGLang is supported too. You will need serious GPU hardware for a model with 320B total parameters, so for most people the Zen or API route is the practical choice. Self-hosting makes sense when you need your code to stay on your own servers.
Limits to know
- The Zen free model is time-limited and only works inside OpenCode.
- Thinking is always on, so simple tasks can use more tokens than you expect. Use low effort for quick jobs.
- The benchmarks above are Z.ai’s own. Check them on your own code.
- Z.ai’s launch blog didn’t load when we checked, so we haven’t confirmed the exact release date.
Want a shorter take? I also cover GLM 5.3 Flash in OpenCode on the AI Profit Boardroom blog.
My verdict on GLM 5.3 Flash OpenCode
If you already use OpenCode, try the free model today. It costs nothing while the window is open and it is strong at real coding tasks. When the window closes, a Z.ai key at $0.15/$0.50 per million tokens is still very cheap. For most people, GLM 5.3 Flash OpenCode is the cheapest way to get near-frontier coding help right now.
About Julian Goldie
I’m Julian Goldie — founder of Goldie Agency and the AI Profit Boardroom. I test new AI agents almost every day for 432K+ YouTube subscribers and run Hermes, OpenClaw, OpenCode, Claude and Codex side by side in my own Agent OS. Claim your free strategy session or join the AI Profit Boardroom.
FAQ: glm 5.3 flash opencode
Is GLM 5.3 Flash free in OpenCode?
Yes, for a limited time. OpenCode Zen offers GLM-5.3-Flash Free (glm-5.3-flash-free), and it only works inside OpenCode.
How much does GLM-5.3-Flash cost on the Z.ai API?
$0.15 per million input tokens, $0.03 cached input and $0.50 per million output tokens. It is not free on the Z.ai API.
What context window does GLM-5.3-Flash have?
1M tokens of context with up to 128K output tokens.
What was ox-alpha?
Ox-alpha was the anonymous name GLM-5.3-Flash used while it was being tested on OpenCode and OpenRouter before launch.
Can I run GLM-5.3-Flash locally?
Yes. The weights are open under MIT and run with vLLM or SGLang, but you need large GPU hardware for a 320B-parameter model.
Ready to put Hermes to work in your business? Join the AI Profit Boardroom for the Agent OS, setup training and daily updates.
Or claim your free strategy session and we’ll map out your setup.
Starting from zero? Get the free AI course and community.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →Get the FREE AI Course + Community →
Related reading
Last updated October 2026. This page is a living guide to glm 5.3 flash opencode — the facts here move fast and it is updated as they do.
