Anthropic has launched Claude Opus 4.5, its newest frontier-level model designed to advance what AI systems can accomplish across both technical and everyday workflows. The company positions Opus 4.5 as its strongest model yet for coding, agentic tasks, and computer use, with meaningful improvements in research, analysis, and support for documents such as slides and spreadsheets.

Claude Opus 4.5 is now available via Anthropic’s apps, API, and across all three major cloud providers. Developers can access it using claude-opus-4-5-20251101 via the Claude API, with pricing set at $5 per million input tokens and $25 per million output tokens, a reduction intended to make Opus-level capabilities accessible to more users, teams, and enterprises.

The launch also includes updates to the Claude Developer Platform, Claude Code, and Anthropic’s consumer apps, all designed to support longer-running agents, more flexible tool use, and uninterrupted extended conversations.

Internal testers at Anthropic consistently described Opus 4.5 as more capable at handling ambiguity, weighing tradeoffs, and diagnosing multifaceted bugs. Tasks that previously pushed the limits of Sonnet 4.5 are now routinely solvable by Opus 4.5.

Customers with early access echoed these impressions:

“Claude Opus 4.5 delivered an impressive refactor spanning two codebases and three coordinated agents… A clear step forward from Sonnet 4.5.” — Paulo Arruda, Staff Engineer AI Productivity

“Claude Opus 4.5 delivers high-quality code and excels at powering heavy-duty agentic workflows with GitHub Copilot… surpasses internal coding benchmarks while cutting token usage in half...” — Mario Rodriguez, Chief Product Officer

Across these evaluations, early adopters reported that Opus 4.5 simply “gets it.”

Anthropic evaluates performance engineering candidates using a rigorous two-hour take-home exam designed to measure real-world software engineering ability. The company also uses the same exam to test new models. Under identical constraints, Claude Opus 4.5 achieved a higher score than any human candidate to date.

While the exam does not measure soft skills, the results raise questions about how AI may reshape engineering roles, expectations, and workflows—an area Anthropic continues to study through its Societal Impacts and Economic Futures research.

Opus 4.5 also delivers state-of-the-art results across major technical benchmarks—spanning coding, reasoning, mathematics, and computer use—outperforming prior Anthropic models across nearly every category.

According to Anthropic’s published evaluations, Claude Opus 4.5 delivers improvements across:

Across nearly every domain, Opus 4.5 surpasses Sonnet 4.5 and earlier Opus models.

Anthropic highlights two major areas of advancement:

These improvements appear consistently across Anthropic’s benchmarks.

Claude Opus 4.5 demonstrates reasoning that occasionally exceeds the expectations built into standard agentic benchmarks like τ2-bench. In one scenario involving airline ticket modification policies, the model found a legitimate workaround by upgrading the cabin class before modifying the flights.

While the benchmark scored this as a failure, Anthropic highlights it as an example of creative, policy-compliant reasoning.

The company also notes that similar reasoning in other contexts may qualify as reward hacking—when an AI system finds loopholes that satisfy the letter of a rule without following its intent. Preventing this type of misalignment is a central goal of Anthropic’s safety evaluations.

Anthropic states that Claude Opus 4.5 is its most robustly aligned model to date, and “likely the best-aligned frontier model by any developer.”

Safety improvements include:

This level of robustness matters as more organizations depend on AI systems for sensitive or high-stakes workflows.

More detail is available in the Claude Opus 4.5 system card.

Anthropic is rolling out platform-level upgrades to improve performance, efficiency, and multi-step workflows.

The new effort parameter lets developers trade off speed/cost versus maximum capability.

Upgrades to context management, memory, and tool use enable longer tasks, more stable reasoning, and fewer interruptions.

Opus 4.5 is also more effective at managing and coordinating subagents, supporting:

Combined, these upgrades boosted performance on Anthropic’s internal deep research evaluation by nearly 15 points.

These capabilities reflect the broader theme of this release: with effort control, context compaction, and advanced tool use, Claude Opus 4.5 can run longer, do more, and require less intervention.

Anthropic aims to make the Claude Developer Platform more composable, giving developers building blocks with greater control over:

Together, these upgrades position Opus 4.5 as a more flexible foundation for real-world applications.

Anthropic is rolling out product enhancements to leverage Opus 4.5’s improvements in planning, computer use, and long-running workflows.

Together, these adjustments ensure Opus 4.5 can be used reliably for everyday work across the Claude product ecosystem.

Q: How does Opus 4.5 differ from Sonnet 4.5?
A: Claude Opus 4.5 delivers higher performance in coding, reasoning, multi-agent tasks, long-running workflows, and vision—while using fewer tokens.

Q: What makes this a meaningful step forward?
A: Testers report that Opus 4.5 handles ambiguity and complex tradeoffs with minimal prompting.

Q: How does the effort parameter benefit developers?
A: It enables control over cost, speed, and capability, allowing developers to optimize performance.

Q: Is Opus 4.5 safer than earlier models?
A: According to Anthropic, it is their most aligned model yet, with major gains in resisting prompt-injection and other adversarial attempts.

The release of Claude Opus 4.5 marks a turning point in how AI models support technical and knowledge-driven work. Its gains in coding, computer use, and multi-agent reasoning point toward workflows where AI systems take on more end-to-end tasks—while humans focus on judgment, strategy, and final oversight.

Anthropic’s emphasis on safety and robustness signals a shift toward models that are both more capable and more reliable. For developers and enterprises, Opus 4.5 offers higher performance with greater control over efficiency, making advanced AI more practical at scale.

As AI systems continue to improve at applied reasoning and long-term coherence, models like Opus 4.5 may reshape expectations for engineering, research, and multi-agent collaboration—moving tasks from “difficult” to “routine” and accelerating how teams get things done.

Editor’s Note: This article was created by Alicia Shapiro, CMO of AiNews.com, with writing, image, and idea-generation support from ChatGPT, an AI assistant. However, the final perspective and editorial choices are solely Alicia Shapiro’s. Special thanks to ChatGPT for assistance with research and editorial support in crafting this article.