Z.ai Confirms Ox Alpha as GLM-5.3-Flash

// 1h agoMODEL RELEASE

Z.ai Confirms Ox Alpha as GLM-5.3-Flash

Z.ai confirmed that Ox Alpha, the anonymous model previewed through OpenCode and OpenRouter, was GLM-5.3-Flash. The model offers a 1M-token context window, native multimodal input, 320B total parameters with 18B active, and MIT-licensed open weights.

// ANALYSIS

GLM-5.3-Flash turns a stealth-model mystery into a serious open-model contender for long-running coding and agent workflows.

–Native image, video, and file understanding expands coding agents beyond text-only repository work
–Sparse plus linear attention targets lower serving costs for million-token contexts
–Vendor-reported benchmarks show major gains over GLM-5.2 in coding, tool use, and automation
–MIT weights improve portability, but the 320B footprint still makes local deployment demanding
–Developers using Ox Alpha should migrate to the official GLM-5.3-Flash model identifier

// TAGS

glm-5.3-flashllmmultimodallong-contextai-codingcoding-agentopen-weightsopen-source

DISCOVERED

1h ago

2026-08-31

PUBLISHED

9h ago

2026-08-31

RELEVANCE

10/ 10

AUTHOR

viveksingh_r

// KEEP READING

More AI developer news from the feed

EXPLORE FULL FEED

LAUNCH1h ago

Claude Security scans code, suggests patches

Anthropic’s official Claude Code plugin maps repositories, models threats, investigates vulnerabilities with multiple agents, and independently verifies findings before producing targeted patch files for review. It runs inside the developer’s session under existing permissions, using the Claude models available to that account.

NEWS2h ago

GPT-5.6 Sol, Claude Fable 5 Split Coding Wins

A months-long side-by-side comparison finds GPT-5.6 Sol stronger for backend work thanks to speed, cost, and security, while Claude Fable 5 delivers better frontend design and first-pass results. The takeaway: model choice increasingly depends on the shape of the task.

TUTORIAL2h ago

tokensavings-pro splits Sol, Luna work

tokensavings-pro is an open-source Codex skill that separates repository execution from architectural reasoning: Luna reads, edits, and tests code while Sol receives a compressed 1–3K-token task packet. Users install the skill, restart Codex, and invoke `$dualmode` or `/dualmode`.