Robert TaFounder & CEO, Clarity · Building Hyper-Personalized Agents

The most important Opus 5 number isn't an eval score

7 Sept 2026 · 1 min

Watch on YouTube

0:53

In this brief commentary, Robert Ta highlights Anthropic's Opus 5 model, noting its strong performance across coding, data analysis, design, biology, and general knowledge work. He emphasizes that the model's most compelling attribute is its exceptional resistance to prompt injection attacks. Citing Loris Cherny and the Grayswan API benchmark, Ta explains that Opus 5 exhibits near‑zero success rates for indirect prompt injection, especially when layered defenses are applied—combining robust model alignment, dedicated prompt‑injection probes, and Claude Code's auto‑mode. He positions this robustness as a unique class advantage, contrasting Opus 5 with competing models such as GPT‑5.6 SOL, which he suggests lags behind in this security dimension.

Chapters

Loris Cherny, creator of Claude Code, says Opus 5 is a great model for coding, data analysis, design, biology, and knowledge work. More than any of these eval scores, what is most exciting to me is something else. Opus 5 is our least prompt-injectable model yet. It is a bit buried in the system card, but across AI evals and red teaming, Opus 5 is very hard to prompt inject successfully. And when layering defenses, strong model alignment combined with prompt injection probes combined with auto mode in Claude Code, the success rate for prompt injection attacks drop to zero or nearly zero. This is very new and exciting. And this is the indirect prompt injection robustness, which is a Grayswan API benchmark. And they show that, hey, we are ahead and our models are in class of its own versus the other models, including GPT-5.6 SOL, which is going to be their competitor to the Stable Diffusion and now 5 tiers of intelligence.