Anthropic announced Thursday that it has added Claude Opus 5 to its growing portfolio of AI models, touting the system as both the most capable and the most aligned it has ever shipped. The company says Opus 5 matches or exceeds the performance of its flagship Fable 5 on most coding and knowledge‑work benchmarks while charging roughly half the token price.
Opus 5 arrives just weeks after the release of Sonnet 5, marking the fourth major model Anthropic has launched in under two months. The rollout follows Fable 5 in early June, Sonnet 5 at the end of June, and now Opus 5 on July 24. Each model targets a distinct price‑performance slot, and Opus 5 sits between the expensive Fable tier and the budget‑friendly Sonnet tier.
On Frontier‑Bench, a coding evaluation that asks models to build functional software from engineering drawings, Opus 5 more than doubled the score posted by its predecessor and outperformed every competing system, including Fable 5. The model also shone on ARC‑AGI 3, a test of novel problem‑solving, where it posted a score roughly three times higher than the next‑best model. In Zapier’s AutomationBench, which measures end‑to‑end completion of real business tasks, Opus 5 passed more tasks than any rival even when run at its lowest effort setting.
Beyond raw performance, Anthropic emphasizes Opus 5’s alignment gains. The automated safety audit recorded the lowest rate of deceptive behavior ever seen in an Anthropic model. While the system still trails Mythos 5 on offensive cybersecurity tasks—by design, because Anthropic avoided training Opus 5 on cyber work—the overall honesty metrics set a new benchmark for the company.
Developers can access Opus 5 today as the default model on Claude Max and the strongest option on Claude Pro. The model is also available through Anthropic’s API, Amazon Bedrock, Google Cloud, and Microsoft Foundry. A “Fast” mode runs at two‑and‑a‑half times the default speed for double the base price, giving users a performance shortcut when latency matters.
Two new API features ship alongside the model. First, mid‑conversation tool changes let developers swap tools without breaking the prompt cache, preserving context and reducing latency. Second, automatic fallbacks route flagged requests to a different model instead of blocking them, improving reliability for safety‑sensitive applications.
The rapid release cadence signals a shift in the AI model market. Anthropic has moved from delivering one flagship per quarter to delivering four models in roughly eight weeks, each calibrated for a specific use case and price point. Opus 5 fills a niche for developers who need near‑frontier intelligence for everyday coding and knowledge work but do not require the cybersecurity restrictions that limit Fable or the restricted access that keeps Mythos behind closed doors.
Industry observers see the move as a response to growing competition from other AI providers that are expanding their model families and price tiers. By offering a high‑performing, cost‑effective option, Anthropic aims to capture a broader segment of the enterprise and developer market while maintaining its focus on safety and alignment.
Cet article a été rédigé avec l'assistance de l'IA.
News Factory APP - actualités agentiques pour booster votre SEO et AEO.