Ox Alpha on OpenRouter identified as GLM-5.3 variant with alignment tax
airesearch12 · x · 2026-08-21
Analysis suggests the stealth model "Ox Alpha" on OpenRouter is a variant of Zhipu's GLM-5.3. While trait differences are minimal, Ox Alpha shows increased alignment and safety features—such as a more clinical register and sanitized output—indicating additional post-training. This "alignment tax" results in a significantly lower ELO score on EQ Bench 4 compared to GLM-5.3, likely a trade-off to mitigate risks for open-weights release. The model remains outstanding despite the drop.
More from Models
- Stefano Ermon teases 'largest and most capable model' landing next week, opens early access — StefanoErmon · 2026-10-16
- Inception teases its biggest model yet after Mercury 2 makes TIME's Best Inventions — StefanoErmon · 2026-10-16
- Diffusion LLM Mercury 2 named to TIME's Best Inventions of 2026 — StefanoErmon · 2026-10-16
- Gemini 4 Argon scores 97.5% on ARC-AGI-3, second lab past 95% milestone — fchollet · 2026-10-16
- New inference engine 'e' hits 10,500 tok/s on a single DGX Station — brandon_galang · 2026-10-16
- NVIDIA's Nemotron 3.5 Super VL hits CoreWeave: 12B-active multimodal MoE at $0.25/M input tokens — wandb · 2026-10-16