Anthropic Releases Claude Opus 4.5: AI Model for Enhanced Productivity

Redazione RHC : 24 November 2025 21:50

Anthropic has released Claude Opus 4.5 , its new flagship model, which the company says is the most powerful version yet and ranks at the top of the class for practical programming, agent-based productivity scenarios.

The model has also seen significant improvements in in-depth search, analytics, and presentation capabilities. Opus 4.5 is now available via apps, APIs, and across all three major cloud technologies.

SOTA in real engineering

In the SWE-bench Verified test, the new model shows the best result among all frontier models : Anthropic particularly emphasizes that Opus 4.5 represents a significant step forward compared to Sonnet 4.5, overcoming tasks that only a few weeks ago were considered “almost impossible” for the previous generation.

Furthermore:

• Opus 4.5 is a leader in 7 out of 8 programming languages on SWE-bench Multilingual.

• Improvements aren’t limited to code: the model has seen significant advances in vision, mathematics, reasoning, and multimodal tasks.

• On Aider Polyglot , BrowseComp-Plus, Vending-Bench – also SOTA indicators or similar.

In the context of artificial intelligence , “SOTA” (State of the Art) refers to the model or technique that achieves the best known performance on one or more relevant benchmarks.

An example of improvement was a case of the τ² benchmark : the model would have ranked seventh, after GPT 5.1.

Stronger, smarter, safer

According to the team, Opus 4.5 is Anthropic’s most secure and resistant to immediate injection. It outperformed all competitors in a series of attack-demand resilience tests. Furthermore:

• Our internal performance review of Opus 4.5 yielded better results than any other test we’ve ever run.
• Thanks to an improved reasoning pipeline, the model uses significantly fewer tokens for reasoning and finding solutions.

Force control, compaction and multi-agent

Opus 4.5 introduces an important new feature for developers: the effort parameter , which determines the depth of reasoning:

• With average effort, the model replicates Sonnet 4.5 using 76% fewer tokens.
• At most, it outperforms Sonnet 4.5 by 4.3 percentage points, generating 48% fewer tokens.

According to Anthropic, this results in a 15% increase in agents’ in-depth research activities.

The cost of Claude Opus 4.5

Claude Opus 4.5 di Anthropic adotta un modello di pricing flessibile, suddiviso tra API e abbonamenti. Per gli sviluppatori che utilizzano l’API, il costo è calcolato per milione di token: la tariffa è di $5,00 per l’input (i dati che invii al modello) e $25,00 per l’output (la risposta generata), rendendolo sensibilmente più economico rispetto al suo predecessore Opus 4.1.

Anthropic offre anche sconti significativi tramite funzionalità come l’elaborazione batch (sconto del 50%) e il caching dei prompt (fino al 90%). Per gli utenti finali che utilizzano l’interfaccia web, Opus 4.5 è accessibile tramite i piani in abbonamento: il piano base è il “Pro” a $20 al mese, che offre l’accesso al modello con un limite di utilizzo giornaliero, mentre il piano “Max” (a partire da $100 al mese) offre un utilizzo notevolmente maggiore.

Platform and product updates

With the release of Opus 4.5 the following updates have been introduced:

• Claude Code: The new Plan mode generates detailed plans, asks any clarifying questions, and creates an editable plan.md file before execution.
• Claude Code is now also available in the desktop app, with support for both local and remote parallel sessions.
• In the Claude app, extended conversations no longer “hang”: the previous context is automatically compressed.
• Claude for Chrome is now available to all Max users.
• Claude for Excel has been expanded into beta for Max, Team, and Enterprise users.

Anthropic has also increased the usage limits for Opus 4.5, making it more suitable as a primary work tool. The company stated that users will receive approximately the same volume of Opus tokens as previously available Sonnet tokens.