Anthropic has released a new Claude model, Opus 5, bringing it closer to the capabilities of Claude Fable 5 at half the price. The new model produces the highest work performance the company has ever offered, with even more gains in its coding capabilities.
In a fight for AI supremacy, Anthropic joins Google and OpenAI, which earlier this month released new AI models, all more efficient and better at certain tasks. Although Fable 5 remains Anthropic’s most “ambitious” work, the latest model excels in business with its coding capabilities and much more.
Anthropic says Opus 5’s strengths lie in knowledge work, autonomy and biology. It is able to work more independently to complete tasks without user intervention and with less switching back and forth than previous models. He can also now double-check his own work and recover from errors as he goes.
The quality of scientific research has increased with Opus 5 compared to Opus 4.8. These improvements are most visible in organic chemistry tasks, such as inferring molecular structures from spectroscopy data, according to Anthropic.
You can now also adjust the effort levels for the model. There is a “dial” that can be adjusted to indicate how hard the AI is thinking about a specific problem or task. New safety and security guardrails have also been put in place, which Anthropic calls the “safest model yet and the hardest to get people to do harmful things.”
Although the model is not specifically designed for cybersecurity, its safeguards make it more useful for tasks like secure coding and vulnerability scanning.
New Quick Mode and more
Claude now has a research preview for an Opus 5 quick mode. Anthropic claims that this mode provides full intelligence of Opus 5 with 2.5x faster output token generation at 2x the standard price of Opus 4. The mode is also available through Claude Code with additional usage credits.
Anthropic introduces a “fallback model” when you request something that triggers a strict security filter, only to block the request. In these cases, the system will automatically switch to another template to attempt to provide a response to the prompt. You can choose which model to rely on in these scenarios.
There is also a new option for developers that will allow them to change the tools Claude can use mid-conversation without invalidating the prompt cache. This means that each phase of an agent’s job is only exposed to the tools they need to complete their task.
Blake has over a decade of experience writing for the web, focusing on mobile, where he covered the smartphone boom of the 2010s and the broader tech scene. When he’s not at a keyboard, you’ll likely find him playing video games or watching horror movies. See full bio
