Ox Alpha (GLM-5.3-Flash) Explained + Free Access | TeamoRouter

Quick Answer

Ox Alpha is GLM-5.3-Flash, Zhipu AI's open-source frontier model — the anonymous identity that quietly landed on OpenRouter and OpenCode on August 20, 2026, hit the top of OpenRouter on day one, set a new single-day usage record, and consumed roughly 62 trillion tokens in six days before Zhipu revealed its identity on August 26.

• Identity: Ox Alpha = GLM-5.3-Flash (the Chinese developer community calls it "牛来" / the "Ox" model).
• Specs: 320B total / 18B active parameters, 92→45 layers, 1.04M-token context, native multimodal (text/image/video/file), MIT open source.
• Performance: AA Intelligence Index 57, tied with Claude Opus 4.8 and ahead of DeepSeek V4 Pro (53).
• Pricing: ¥0.8 in / ¥2.8 out per 1M tokens domestically (about 1/10th of GLM-5.3); $0.3 in / $1.2 out internationally (about 1/40th of Claude Opus 4.8).
• Free access: TeamoRouter's welfare tier glm-5.3-flash-free gives you 200 free requests per day — register, grab a key, and call it over the OpenAI-compatible API. Paid glm-5.3-flash has no daily cap.

The Six-Day Blind Test That Broke OpenRouter

Ox Alpha didn't arrive with a press release. It was dropped onto OpenRouter and OpenCode on August 20, 2026 as an anonymous model — no company name, no architecture notes. Developers voted with their traffic:

• Day one: shot to #1 on OpenRouter and set a platform single-day usage record at 4x the previous peak.
• OpenCode: ended DeepSeek's 56-day run at the top of the board.
• Six days (Aug 20–26): roughly 62 trillion tokens consumed globally, about 20% of OpenRouter's weekly traffic, more than 2x DeepSeek's usage.
• August 26: Zhipu revealed the identity — Ox Alpha is GLM-5.3-Flash, open-sourced under MIT that same evening. Zhipu's stock (02513.HK) rose 12.62% and its market cap crossed back above 500 billion HKD.

Why "Ox"? The codename means "bull/ox" in English, and a film called 《牛来》 was trending in China at the same time — so Chinese developers started calling it the "牛来 model" (the "bull comes" model). In Chinese technical communities, "牛来模型" and "Ox Alpha" now refer to the same thing.

Hard Specs at a Glance

Three takeaways:

1. It's a MoE model — 320B total, 18B active. Flagship knowledge capacity at a fraction of the per-token cost.
2. 92→45 layers — the leaner stack plus hybrid attention is what makes the 1.04M-token context affordable to serve.
3. First of its kind — it's the first open-source frontier model to fuse sparse attention with linear attention, adding a new manifold-constrained hyperconnection (mHC).

Performance: Tied with Claude Opus 4.8

On the Artificial Analysis Intelligence Index, GLM-5.3-Flash scores 57 — the same as Anthropic Claude Opus 4.8, and above DeepSeek V4 Pro's 53. That puts it at the top of the open-source tier for overall intelligence, which is a big deal at its price point.

It also opened a new chapter for Chinese hardware: all inference traffic runs on a 100,000-chip domestic cluster, Zhipu's first large-scale deployment of domestic chips for inference. That's both a supply-chain statement and a big part of why the price can be so low.

Pricing: Frontier Intelligence, Commodity Price

Domestic pricing (per 1M tokens):

That's about 1/10th of GLM-5.3, and about 1/20th during the limited-time half-price window.

International pricing (per 1M tokens):

That's about 1/40th of Claude Opus 4.8's official price, and Zhipu says the overall bill is even lower than the adjusted DeepSeek V4-Flash. In short: open-source model, frontier intelligence, commodity price.

How It Compares: DeepSeek V4 and Claude Opus 4.8

• vs DeepSeek V4 Pro: GLM-5.3-Flash leads on the AA Intelligence Index (57 vs 53). During its anonymous week it pulled more than 2x DeepSeek's traffic on OpenRouter. Both sit at "commodity cheap" pricing, so the choice comes down to intelligence and multimodal capability, where GLM edges ahead.
• vs Claude Opus 4.8: identical AA score (57) — that's the headline. The difference is the bill: the international price is about 1/40th of Opus 4.8's official rate, and GLM-5.3-Flash is MIT open source, so you can self-host the same intelligence instead of renting it.

One-line summary: closed-source flagship intelligence at open-source prices. For Chinese developers, that math — combined with no-proxy access and Alipay billing through a gateway like TeamoRouter — makes it the easiest frontier-class model to actually use at scale.

Free Access: 3 Paths to GLM-5.3-Flash

You can use GLM-5.3-Flash without paying, three ways:

For most developers, the TeamoRouter welfare tier is the fastest free path: no GPU, no overseas card, no grabbing experience cards — and it's a real API you can call from code.

How to Get Free Access on TeamoRouter

1. Register and create a key

Sign up at TeamoRouter and create an API key (sk-teamo-...). China-friendly: Alipay top-up, no proxy needed, failed requests aren't billed — and the free tier costs nothing.

2. Confirm the model exists

Pull the live model list:

curl https://api.teamorouter.cn/v1/models \
  -H "Authorization: Bearer sk-teamo-your-key"

Look for glm-5.3-flash, glm-5.3-flash-free (welfare), glm-5.3, and glm-5.2.

3. Call it (OpenAI-compatible)

GLM uses the OpenAI-compatible endpoint POST /v1/chat/completions with base URL https://api.teamorouter.cn/v1. Note: /v1/responses is GPT-series only — GLM can't use it.

cURL:

curl https://api.teamorouter.cn/v1/chat/completions \
  -H "Authorization: Bearer sk-teamo-your-key" \
  -H "content-type: application/json" \
  -d '{
    "model": "glm-5.3-flash-free",
    "max_tokens": 1024,
    "messages": [{"role": "user", "content": "Write a Python quicksort"}]
  }'

Python (openai SDK):

from openai import OpenAI

client = OpenAI(
    api_key="sk-teamo-your-key",
    base_url="https://api.teamorouter.cn/v1",
)

resp = client.chat.completions.create(
    model="glm-5.3-flash-free",
    messages=[{"role": "user", "content": "Write a Python quicksort"}],
)
print(resp.choices[0].message.content)

Node:

import OpenAI from "openai";

const client = new OpenAI({
  apiKey: "sk-teamo-your-key",
  baseURL: "https://api.teamorouter.cn/v1",
});

const resp = await client.chat.completions.create({
  model: "glm-5.3-flash-free",
  messages: [{ role: "user", content: "Write a Python quicksort" }],
});
console.log(resp.choices[0].message.content);

4. Point Claude Code / Codex at it

One TeamoRouter key speaks both Anthropic and OpenAI protocols, so you can run GLM-5.3-Flash inside Claude Code:

export ANTHROPIC_BASE_URL="https://api.teamorouter.cn"
export ANTHROPIC_AUTH_TOKEN="sk-teamo-your-key"
export ANTHROPIC_MODEL="glm-5.3-flash-free"
export ANTHROPIC_SMALL_FAST_MODEL="glm-5.3-flash-free"

Or Codex / other OpenAI-protocol tools:

export OPENAI_BASE_URL="https://api.teamorouter.cn/v1"
export OPENAI_API_KEY="sk-teamo-your-key"

When the Free Quota Runs Out

Two fallbacks, same key, no re-registration:

1. Switch to paid glm-5.3-flash — change only the model field. No daily request cap, and the price is still very low (¥0.8/¥2.8 per 1M domestic).
2. Fall back to DeepSeek welfare tiers — TeamoRouter also offers deepseek-v4-flash-free (200/day) and deepseek-v4-pro-free (50/day). GLM and DeepSeek quotas are separate, so you effectively get double free buckets from one key.

FAQ

Is Ox Alpha the same as GLM-5.3-Flash?

Yes. Ox Alpha was the anonymous blind-test identity from August 20; Zhipu confirmed on August 26 that it is GLM-5.3-Flash.

Is GLM-5.3-Flash open source?

Yes — MIT license, released August 26, 2026. You can self-host it.

Is glm-5.3-flash-free really free, 200 times a day?

Yes. TeamoRouter's welfare-tier rule: models ending in -free get a daily request quota — glm-5.3-flash-free is 200 requests per day (RPD), reset daily.

Can GLM use /v1/responses?

No. /v1/responses is GPT-series only. Use /v1/chat/completions for GLM.

Do I need a GPU or an overseas card?

No. The TeamoRouter welfare tier needs neither — just a sk-teamo- key.

Bottom Line

Ox Alpha / GLM-5.3-Flash is the model event of August 2026: a six-day anonymous blind test that broke OpenRouter records, frontier-level intelligence (AA 57) at a fraction of flagship prices, open-sourced under MIT, and running on a 100K domestic-chip cluster. The fastest way to try it — for free — is TeamoRouter's welfare tier, glm-5.3-flash-free, at 200 requests per day. One key routes GLM alongside Claude, GPT, DeepSeek, Kimi, Gemini, and Grok, with unified billing.

Get your TeamoRouter key and claim 200 free GLM-5.3-Flash requests a day, then wire it into Claude Code, Codex, or any OpenAI-compatible tool.