Ox Alpha and GLM-5.3-Flash: What Is Known and Unverified

Z.ai describes GLM-5.3-Flash as the first native multimodal model in its GLM-5 series. Its official documentation lists 320 billion total parameters, 18 billion activated parameters, input for video, images, text, and files, a 1M-token context length, and a maximum output of 128K tokens. It produces text. These are specifications for GLM-5.3-Flash; available sources do not establish whether the “Ox Alpha” label referred to this checkpoint or to a Z.ai test deployment.

What Z.ai documents about GLM-5.3-Flash

The architecture combines sparse and linear attention. Z.ai reports that this reduces attention computation by 3.01 times and KV cache by 4.44 times compared with GLM-5.3. Those are provider comparisons and should be read in that context. Z.ai describes the model as multimodal visual coding capable, with image and video inputs, but its API documentation specifies text output.

The API model called FlashX is a separate offering. The current pricing page lists it at $0.37/M input, $0.075/M cached input, and $1.25/M output. The coding-plan quota system is also distinct from per-token API billing: Z.ai says off-peak plan calls, including weekends, consume half the standard points. That discount does not mean all API token prices are universally discounted by 50%.

What remains unknown about Ox Alpha

The cited GLM documentation and release page describe GLM-5.3-Flash, not the history of the Ox Alpha alias. No official attribution or reproducible evidence in these sources identifies Ox Alpha as a Z.ai deployment, benchmark checkpoint, or stealth preview. Its provenance should remain unconfirmed until such evidence is available.

How to evaluate the model

When comparing GLM-5.3-Flash with another model, use results from the same benchmark version, harness, tools, and budget. A terminal-agent score does not by itself establish performance on repository issue resolution, and a provider-reported result is not an independent replication. Also account for both input and output volume: a low per-token price alone does not predict the bill for a long agent run.

Read the current Z.ai model documentation and pricing schedule before integrating. Browse our AI Model Leaderboard for RankLLMs’ maintained comparison data.