Ox Alpha Is GLM-5.3-Flash: Data Terms and Price

Ox Alpha model was revealed as Z.ai's GLM-5.3-Flash

OpenRouter says Ox Alpha was developed and operated by Z.ai and revealed to be GLM-5.3-Flash. A banner at the top of the Ox Alpha page states it, and the page for GLM 5.3 Flash answers yes to the question "Was ZAI GLM-5.3 Flash the stealth model Ox Alpha?"

A stealth model is a model released without naming its developer. Before the reveal, the Ox Alpha page described a third-party provider that had chosen to stay anonymous during the preview. OpenRouter routes requests to Ox Alpha and is not its developer, owner or provider.

Ox Alpha was released on August 20, 2026, and GLM 5.3 Flash on August 26, 2026, according to the two pages. The two dates are 6 days apart. For GLM 5.3 itself, see GLM 5.3 benchmarks, pricing, and open weights.

What the Ox Alpha model offers: 1M context and video input

Ox Alpha takes text, images and video as input, returns text, and has a 1,048,576-token context window. It is a reasoning model aimed at coding, long-running agent work and production workloads.

A context window is how much text the model can take in at once, counting both the prompt and the answer. A token is a small chunk of text, roughly a word or part of one. OpenRouter shows 1,048,576 tokens as 1.0M.

The GLM 5.3 Flash page adds that the model supports up to 1,048,576 completion tokens, the same number as the context window. The model accepts tools for function calling and can return structured outputs that follow a JSON schema.

OpenRouter credits a hybrid sparse and linear attention design with keeping long-context behavior accurate while reducing compute overhead. "Stated specifications of Ox Alpha and GLM 5.3 Flash" gathers the stated values.

Stated specifications of Ox Alpha and GLM 5.3 Flash

Item	Stated value
Context window	1,048,576 tokens
Maximum completion	1,048,576 tokens
Input	Text, images, video
Output	Text
Tool calling	Supported
Structured outputs	Supported, by JSON schema

Prompts sent to Ox Alpha were retained by the provider

OpenRouter's Ox Alpha page says prompts and completions were retained by the provider and are not used for training. A prompt is what a user sends to the model, and a completion is what the model sends back. The retention covers what users sent during the stealth period.

The page ties all other use to the Stealth Model Terms, a document it links to. OpenRouter also states that it routes requests and is not the developer, owner or provider.

The reveal gives the provider a name. The Ox Alpha page does not say what happened to the retained data after the reveal.

The Stealth Model Terms are a separate web page whose wording can change over time. A team that needs to show which terms applied on a given day can keep a dated copy of the whole page as Markdown, next to a copy of the OpenRouter page that links to it.

A dated copy also lets two versions of the page be compared later. To make that copy, try the tool that converts a page URL into Markdown.

Ox Alpha pricing: $0.036 per 1M input tokens

OpenRouter lists GLM 5.3 Flash at $0.036 per 1M input and $0.90 per 1M output. The cache read rate, the price for text the provider has already stored from an earlier request, is $0.0108 per 1M tokens. These figures are for the revealed model: the Ox Alpha page lists no price, and neither page states a free period for Ox Alpha.

The headline input price is the lowest among the 29 providers that serve GLM 5.3 Flash on OpenRouter, and prices differ by provider. Z.ai lists $0.15 per 1M input and $0.50 per 1M output, about 4 times the lowest input price. The lowest input price comes from Open Inference, whose output price is $0.90.

GLM 5.3 Flash prices per 1M tokens on OpenRouter

Price point	Input	Output
Headline price (lowest input, Open Inference)	$0.036	$0.90
Z.ai	$0.15	$0.50
GLM 5.3 FlashX, the faster variant	$0.37	$1.25

OpenRouter's API is OpenAI-compatible. Because prices differ by provider, the cost of a call can depend on which provider serves it.

Ox Alpha benchmarks: Artificial Analysis and Vals AI scores

OpenRouter's GLM 5.3 Flash page shows scores from Artificial Analysis, Design Arena and Vals AI, from 92.0% on SWE-bench to 0.0% on ProgramBench. The page attributes each score to a named evaluator. The chart plots six selected scores.

Selected GLM 5.3 Flash scores shown on OpenRouter

Source: OpenRouter, GLM 5.3 Flash page.

Artificial Analysis indexes

Index values, not percentages.

41.8Intelligence Index (Artificial Analysis)

71.5Coding Index (Artificial Analysis)

0204060

Benchmark scores

Percent of questions or tasks solved.

91.2%GPQA Diamond (Artificial Analysis)

92.0%SWE-bench (Vals AI)

0.0%ProgramBench (Vals AI)

0.8%SRE Bench (Vals AI)

0255075%

The model scores 91.2% on GPQA Diamond (Artificial Analysis) and 92.0% on SWE-bench (Vals AI). On two other Vals AI tests it scores 0.0% on ProgramBench and 0.8% on SRE Bench. A single score does not show how the model handles a given job, so start with the test closest to that work.