Jailbroken System Prompt and Fingerprinting Link 'ox-alpha' to Zhipu GLM
eyishazyer · x · 2026-08-22
A jailbreak of the ox-alpha model revealed a system prompt confirming an "undisclosed organization." Technical fingerprinting shows its tokenizer aligns almost exactly with GLM-5.3, and its video encoder spends tokens identically to GLM-5V-Turbo. On the DeepSWE benchmark, ox-alpha currently outperforms Claude Fable 5 and GPT-5.6-Sol, with multiple technical signals pointing to Zhipu AI.
More from Models
- Paying for both Fable and Astra, users still can't easily reach Gemini's best models — dreamwieber · 2026-10-16
- Baseten serves Meta's Muse Spark Open on day 0 at $1.25/M input tokens — baseten · 2026-10-16
- Anthropic red team's first-ever model "abliteration" cut GLM-5.3 refusals from 90% to 3% for ~$4,400 — cephaloform · 2026-10-16
- Meta releases Muse Spark Open: 816B MoE, 42B active, 1M context, day-0 on Baseten — baseten · 2026-10-16
- Developer to Google: loosen usage limits and merge your AI apps into one — dreamwieber · 2026-10-16
- Meta releases 816B-A42B open-weight Muse Spark Open; Apple MLX optimized for it — BenBajarin · 2026-10-16