Anthropic introduced Claude Opus 5 on Friday, positioning the new model as a cost‑effective option for coding, autonomous agents and enterprise workflows.
Pricing and model hierarchy
The model is priced at $5 per million input tokens and $25 per million output tokens, the same rates as Opus 4.8. It becomes the default on Claude Max, Anthropic’s premium consumer tier, and the top offering on Claude Pro. While the company still lists Claude Fable 5 as its most capable system, Opus 5 is marketed as the “daily driver” for complex tasks that can be reviewed after completion.
According to an Anthropic spokesperson, the lineup now stratifies: “Fable 5 for your most ambitious work, the days‑long autonomous projects … Sonnet 5 for work you run at scale, where speed and cost per call decide what ships. Haiku 4.5 for subagents and instant answers.”
Related: Understanding Multi‑Turn AI Attacks that Breach Models 88% of the Time (2026 Guide)
Benchmark performance
Anthropic reports that Opus 5 sets new state‑of‑the‑art marks on several evaluations. On Frontier‑Bench v0.1, a coding benchmark, it scored 43.3 percent—more than double Opus 4.8’s 18.7 percent and ahead of Fable 5’s 33.7 percent—while using less compute per task.
The company notes that Opus 5 lags behind Mythos 5 on cybersecurity and biology tasks, and an OpenAI model still leads on one agentic coding benchmark. The spokesperson added, “The evals where Opus 5 wins are bounded tasks with a specific outcome… What those evals don’t measure is duration.” He suggested running both a bounded task and a long‑horizon job to compare Opus 5 and Fable 5.
Safety and fallback mechanisms
Anthropic says Opus 5 is its most aligned model to date, scoring 2.3 on overall misalignment—a lower figure than Opus 4.8, Sonnet 5 or Fable 5. The company deliberately omitted cybersecurity training, yet the model still identified vulnerabilities at a 79.4 percent rate on Anthropic’s OSS‑Fuzz benchmark, close to Mythos 5’s 80 percent. However, it succeeded in only four exploit challenges versus Mythos 5’s thirteen, reflecting a design choice to limit offensive capabilities.
Related: First Study Links Cyber Fraud Convergence
When a safety classifier intervenes, requests fall back to Opus 4.8, which the spokesperson said reduces the risk of harmful use because the fallback model has lower capability. Users receive a visible message indicating the fallback.
On scientific tasks, Opus 5 outperformed Opus 4.8 on the company’s internal chemistry benchmark by 10.2 percentage points, though Mythos 5 remains stronger for long‑horizon drug‑design projects.
Anthropic’s commercial context includes a recent $380 billion valuation and a projected $9 billion in annualized revenue by the end of 2025. The company has secured a $30 billion Azure compute deal alongside agreements with Google Cloud and Nvidia, highlighting the importance of models that can deliver more work per token.
Related: Westfalia merges luxury RV with compact camper van
Regulatory developments also affect the market. A U.S. judge approved a $1.5 billion copyright settlement with authors, and the U.S. government moved to block foreign access to Anthropic’s most advanced models, highlighting export‑policy considerations.
Additional features released Friday include a “Fast” mode operating at roughly 2.5 times default speed for twice the base price, automatic fallback routing on the API, and mid‑conversation tool changes that preserve the prompt cache. The model carries zero data‑retention requirements for general access, a point the spokesperson emphasized for customers with strict privacy needs.
Developers can access Opus 5 via the Claude API under the identifier claude‑opus‑5. Whether the model’s efficiency claims hold up under large‑scale production workloads remains to be seen, but Anthropic’s focus on everyday utility over peak performance signals a notable shift in the AI sector’s competition.
