Detailed Analysis
Anthropic released Claude Opus 4.7 on April 16, 2026, marking a substantive generational upgrade to the Opus 4.6 model with measurable advances across coding, vision processing, and multi-step reasoning. The new model demonstrates particular strength in software engineering tasks that previously required human oversight, including the resolution of complex race conditions and improved agentic execution across multi-step workflows. Benchmarking results show Opus 4.7 passing three TBench tasks that prior models failed entirely, and outperforming competing frontier models including GPT-5.4 and Gemini 3.1 Pro across finance agent evaluations, GDPval-AA assessments, and professional knowledge work tasks. Pricing remains unchanged at $5 per million input tokens and $25 per million output tokens, with the model available across Claude's consumer tiers (Pro, Max, Team, Enterprise) as well as the API and major cloud platforms including Amazon Bedrock, Google Vertex AI, Microsoft Foundry, and GitHub Copilot.
Among the most technically notable improvements is a tripling of visual input resolution capacity, with Opus 4.7 now processing images up to 2,576 pixels on the long edge — a change that substantially broadens the model's utility for vision-intensive professional applications such as document analysis, medical imaging review, and technical diagram interpretation. The release also introduces several new developer-facing features: an "xhigh" effort level designed to let developers balance reasoning depth against latency requirements, task budgets entering public beta, and an `ultrareview` command within Claude Code specifically designed to surface software bugs. A new tokenizer, producing 1.0–1.35 times more tokens than its predecessor, suggests architectural-level changes to how the model processes and encodes information.
The release arrives alongside notable safety disclosures that illuminate the internal tradeoffs Anthropic is navigating. While Opus 4.7 includes cyber safeguards designed to block high-risk uses, Anthropic has simultaneously acknowledged the existence of a more capable model — internally referred to as "Mythos" — whose broader release has been restricted due to safety concerns addressed under an initiative called Project Glasswing. Mythos is currently available only to vetted security professionals through a Cyber Verification Program, underscoring Anthropic's stated commitment to staged and risk-stratified deployment of frontier capabilities. This dual-track approach — releasing a broadly capable but constrained public model while maintaining a more powerful, access-restricted version — reflects a deliberate strategy to extract commercial value from advanced AI systems while managing potential misuse vectors.
The Opus 4.7 launch fits within a broader competitive pattern in the frontier AI market, where the cadence of major model releases has accelerated and differentiation increasingly depends on agentic and long-horizon task performance rather than raw language generation quality alone. Anthropic's emphasis on coding benchmarks and software engineering capabilities speaks directly to the enterprise developer market, where Claude Code and API integrations with platforms like GitHub Copilot position the model as a direct productivity tool rather than a general-purpose assistant. The inclusion of task budgets in public beta further signals Anthropic's intent to give enterprise customers fine-grained cost and compute controls — a feature set increasingly essential as agentic deployments generate variable and sometimes unpredictable token volumes. Collectively, the Opus 4.7 release demonstrates Anthropic advancing on multiple simultaneous fronts: model capability, developer tooling, safety architecture, and enterprise distribution.
Read original article →