Photo: IMEN BEN YOUSSEF / AFP / Getty Images
On Tuesday (September 22), Anthropic announced the release of its latest AI model, Opus 5.5, which the company claims is its "safest" yet. This new model is designed to be faster and cheaper than its predecessors, offering a 40% reduction in operational costs. The model is part of Anthropic's Claude lineup, with Opus being the most advanced tier.
According to TechCrunch, Opus 5.5 is priced at $20 per million output tokens, a significant decrease from the previous version. The model also boasts improved efficiency, using fewer tokens to complete tasks and generating output 30% faster.
The model's release comes with enhanced safety features, making it less likely to perform actions that cannot be undone or to act outside its given parameters. Opus 5.5 is designed to avoid answering questions related to hacking, biology, and AI research, especially those flagged for misuse. These safeguards align with Anthropic's commitment to pacing AI capability advancements with risk prevention measures, as highlighted by CEO Dario Amodei in a recent blog post.
In addition to its safety features, Opus 5.5 outperforms previous models on several benchmarks. It scored 66.4% on Terminal-Bench 4.0 and 54.4% on FrontierCode, surpassing its predecessor, Fable 5.1, in these areas. The model's performance in real-world coding tasks has also been impressive, with early testers reporting significant reductions in task completion time and token usage.
Anthropic plans to release cheaper versions of Opus 5.5 in the coming weeks, further expanding its AI offerings. As the company continues to enhance its models, it emphasizes the importance of public policy in ensuring AI systems' safety and reliability. More details on these efforts are expected to be shared soon.