Claude Opus 4.7 Arrives: Anthropic’s Latest Frontier Model Sets New Standards in Coding, Reasoning, and Creative Work

On April 16, 2026, Anthropic announced the general availability of Claude Opus 4.7, the latest iteration in its flagship Opus model family. The release marks a significant step forward in AI capabilities — particularly in advanced software engineering, long-context reasoning, and creative professional tasks — while also serving as an important testing ground for safety measures ahead of the eventual broader release of the even more powerful Claude Mythos Preview model.

A Measurable Leap in Coding Performance

The headline improvement in Opus 4.7 is its substantially stronger performance on complex coding tasks. On Anthropic’s internal 93-task coding benchmark, the new model improved its resolution rate by 13% over Opus 4.6. More notably, it successfully completed four tasks that neither Opus 4.6 nor Claude Sonnet 4.6 could solve — demonstrating genuine capability expansion rather than marginal improvement.

Early users are already reporting transformative changes in their workflows. Developers describe being able to hand off their most difficult coding work — the kind that previously required close, hands-on supervision — to Opus 4.7 with genuine confidence. The model handles complex, long-running tasks with rigor and consistency, pays precise attention to instructions, and has developed the ability to verify its own outputs before reporting back to users.

Combined with faster median latency and stricter instruction following, these improvements are particularly meaningful for developers working on multi-step engineering workflows. As one industry observer noted: “Opus 4.7 cuts the friction from those multi-step tasks so developers can stay in the flow and focus on building.”

Research Agent Benchmark: Market-Leading Efficiency

Opus 4.7’s capabilities extend well beyond coding. On Anthropic’s internal research-agent benchmark — which tests multi-step analytical work across six modules — Opus 4.7 tied for the top overall score at 0.715. The model also delivered the most consistent long-context performance of any model Anthropic has tested to date.

The gains are particularly pronounced in specialized domains. On the General Finance module — the largest of the six test modules — Opus 4.7 scored 0.813 compared to Opus 4.6’s 0.767, representing a meaningful improvement in financial reasoning and analysis. The model also showed the best disclosure and data discipline in the group, and demonstrated stronger performance on deductive logic tasks.

“Claude Opus 4.7 extends the limit of what models can do to investigate and get tasks done,” said an industry evaluator. “Anthropic has clearly optimized for sustained reasoning over long runs, and it shows with market-leading performance. As engineers shift from working 1:1 with agents to managing them in parallel, this is exactly the kind of frontier capability that unlocks new workflows.”

Enhanced Vision and Creative Capabilities

Beyond reasoning and coding, Opus 4.7 brings substantially improved vision capabilities. The model can now process images at higher resolution, enabling it to extract more detail and nuance from visual inputs — a critical improvement for tasks ranging from diagram analysis to medical imaging review.

The model also demonstrates notable gains in creative and design-oriented tasks. Anthropic reports that Opus 4.7 is “more tasteful and creative when completing professional tasks,” producing higher-quality user interfaces, presentation slides, and documents. This improvement signals a broader trend: frontier AI models are increasingly capable of work that requires aesthetic judgment alongside technical competence.

The Mythos Preview Context: Why Opus 4.7 Matters Beyond Its Own Capabilities

While Opus 4.7 is impressive in its own right, it also plays a crucial role in Anthropic’s broader AI safety strategy. The model is deliberately less capable than Claude Mythos Preview — Anthropic’s most powerful unreleased model — particularly in cybersecurity domains.

During Opus 4.7’s training, Anthropic experimented with efforts to differentially reduce its cyber capabilities. The result is a model whose cybersecurity abilities are not as advanced as those of Mythos Preview. Opus 4.7 is being released with safeguards that automatically detect and block requests indicating prohibited or high-risk cybersecurity uses.

Mythos Preview itself is the centerpiece of Project Glasswing, Anthropic’s cybersecurity initiative announced just days before the Opus 4.7 launch. Project Glasswing brings together Amazon Web Services, Anthropic, Apple, Broadcom, Cisco, CrowdStrike, Google, JPMorganChase, the Linux Foundation, Microsoft, NVIDIA, and Palo Alto Networks to secure the world’s most critical software.

Mythos Preview has already identified thousands of high-severity vulnerabilities, including some in every major operating system and web browser. Anthropic is committing up to $100 million in usage credits for Mythos Preview across defensive security efforts, plus $4 million in direct donations to open-source security organizations. Access has been extended to over 40 organizations that build or maintain critical software infrastructure.

Opus 4.7 serves as the first testing ground for the cyber safeguards that Anthropic plans to eventually deploy with Mythos-class models. What the company learns from real-world deployment of these safeguards on Opus 4.7 will inform the safe, broad release of its most capable models.

Security Professionals: The Cyber Verification Program

For legitimate security professionals who want to use Opus 4.7 for vulnerability research, penetration testing, and red-teaming, Anthropic has launched a new Cyber Verification Program. This initiative provides authorized access to the model’s cybersecurity capabilities while maintaining guardrails against misuse.

Availability and Pricing

Claude Opus 4.7 is available today across all Claude products and the Claude API, as well as through major cloud platforms including Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry. Developers can access the model using the claude-opus-4-7 model identifier via the Claude API.

Pricing remains unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens. Anthropic has maintained this price point despite the model’s significant capability improvements, making advanced reasoning and coding more accessible to developers and enterprises.

What This Means for Developers and Enterprises

The release of Opus 4.7 has several practical implications for organizations already using Claude:

Software teams can now delegate more complex, multi-step engineering tasks to AI agents with greater confidence, reducing the need for close human supervision.
Research analysts benefit from stronger long-context performance and improved financial reasoning, making Opus 4.7 better suited for deep analytical workflows.
Design professionals can leverage enhanced creative capabilities for interface design, presentation creation, and document generation.
Security teams should consider joining the Cyber Verification Program to access the model’s cybersecurity features for legitimate defensive work.

The shift from working one-on-one with AI agents to managing multiple agents in parallel represents a fundamental change in how engineering teams will operate. Opus 4.7 is optimized precisely for this new paradigm — sustained, reliable reasoning over long runs with minimal human intervention.

The Road Ahead

Opus 4.7 is a significant milestone, but it’s also a stepping stone. Anthropic’s commitment to $100 million in cybersecurity usage credits and the ongoing Project Glasswing partnership suggest that the company is preparing the ecosystem for even more capable models. The deliberate, measured approach to releasing Mythos-class capabilities — testing safeguards on Opus 4.7 first — reflects Anthropic’s safety-first philosophy.

As AI models continue to advance at a rapid pace, Opus 4.7 demonstrates that meaningful progress is still being made not just in raw capability, but in the practical qualities that matter most to working professionals: thoroughness, consistency, instruction following, and the ability to self-verify. These may seem like incremental improvements, but in practice, they represent the difference between an AI tool that’s interesting to experiment with and one that’s genuinely reliable enough to put production workloads on.

For developers, researchers, and enterprises looking to leverage frontier AI capabilities today, Claude Opus 4.7 is available now — and it’s arguably the most capable Opus model Anthropic has released to date.

Key Takeaways

Coding benchmark: 13% improvement in resolution rate over Opus 4.6 on a 93-task benchmark; solved 4 tasks no previous model could complete.
Research agent performance: Top-tier score of 0.715 across six modules; 0.813 on General Finance (up from 0.767).
Vision: Higher-resolution image understanding for more detailed visual analysis.
Pricing: Unchanged at $5/M input tokens, $25/M output tokens.
Availability: Claude products, API, Amazon Bedrock, Google Vertex AI, Microsoft Foundry.
Safety: First model to test cyber safeguards ahead of Mythos Preview’s broader release.

Claude Opus 4.7 Arrives: Anthropic’s Latest Frontier Model Sets New Standards in Coding, Reasoning, and Creative Work