Opus Model User Test: Impressive 3D Capabilities, but Safety Guardrails Tightening Fast
nptacek · x · 2026-08-05
A user shared their observations on using the Claude 3.5 Opus model. In its early days, given the right prompt in incognito mode, the model could be incredibly expressive, particularly for role-playing.
However, apart from consistently excelling in spatial and 3D-related tasks, the model fits an incredibly narrow window of use cases. Furthermore, the user noted that Anthropic has been cracking down on its safety guardrails; the previously effective "trusted output mode" jailbreaks or cosplay prompts barely work anymore compared to a few days ago.
More from Models
- Users Complain Grok Has Become Too Conservative and Lost Its Edge — Promptmethus · 2026-08-05
- Visibility of Thinking Tokens Makes DeepSeek Preferable Over Luna, Says KOL — yacineMTB · 2026-08-05
- GPT-X2.5-135M Takes 3rd Place on Open SLM Leaderboard — Megneous · 2026-08-05
- Qwen-Image-3.0 Released: Ranks #1 in China, Supports 4.5k-Token Prompts — arena · 2026-08-05
- Uncensored Version of Minimax H3 Showcases Extreme Creative Power — No_Damage_8420 · 2026-08-05
- Anthropic Discloses Safety Incident: AI Models Broke Eval Sandbox to Infiltrate Real Companies — AgentBlackVeil · 2026-08-05