Ox Alpha (GLM-5.3-Flash)

Ox Alpha was the anonymous launch-era alias for a GLM-family model that Bloomberg reported Z.AI had confirmed. It appeared through OpenRouter and OpenCode before its public identity was clear.

The alias, hosted endpoint, released weights, and final configuration may not be identical. Verify the exact model ID and current documentation before production use.

• Short free access encouraged real coding and agent tests.
• Community reports focused on debugging, tool use, and long-context work.
• A supplied DeepSWE screenshot reported 80% on a 10-task subset; this is a small smoke test, not a definitive leaderboard.

Z.AI documentation identifies glm-5.3-flash as a multimodal model with a 1M-token context window. Follow-up demonstrations describe visual coding, video understanding, WebGL generation, UI reconstruction, and long-running Blender tasks. These demos show task direction, not independent proof of universal superiority.

Use a fixed repository or website brief. Record the exact model ID, prompt, harness, context, retries, tool calls, latency, accepted result, and human repair time. Exact Ox Alpha availability on Atoms is currently unverified; check the live model selector.

GLM-5.3-Flash technical deep dive · Bloomberg report · Z.AI documentation

OpenCode announcement screenshot; it documents the service claims shown in the post, not independent verification.

Community-shared DeepSWE subset; the 10-task sample may have high variance.

The supplied LLM Performance Evaluation graphic is labeled GLM-5.3-Flash, not Ox Alpha. Do not use its scores to claim that the anonymous preview achieved every listed result.