Anthropic launched Opus 4.7 this Thursday. The model represents the latest update to the Opus line and is now available to users on paid plans on claude.ai. Ele can also be accessed via API and through partners such as Amazon Bedrock, Google Vertex AI and Microsoft Foundry.
The launch maintains the same price as the previous version. Opus 4.7 brings concrete advances in tasks that require multiple steps, image analysis and sustained reasoning. Desenvolvedores report that the model requires less intervention in complex programming work.
Detalhes technicians defining Opus 4.7
The Opus 4.7 ranks as the most capable model available overall in the Claude family. Ele is below Claude Mythos Preview, a version kept in a restricted environment for security reasons.
Anthropic designed the update to deliver practical gains in real work scenarios. The model maintains a context window of 1 million tokens and supports up to 128 thousand output tokens. The reliable knowledge cutoff reaches January 2026.
Usuários who test the model on agentic flows notice greater consistency over long tasks. Opus 4.7 checks its own results before completing responses in complex operations.
Introducing Claude Design by Anthropic Labs: make prototypes, slides, and one-pagers by talking to Claude.
— Claude (@claudeai) April 17, 2026
Powered by Claude Opus 4.7, our most capable vision model. Available in research preview on the Pro, Max, Team, and Enterprise plans, rolling out throughout the day. pic.twitter.com/2BgBGtgYGX
Multimodal Visão with increased resolution
One of the most visible changes is in image processing. Opus 4.7 accepts files up to 2,576 pixels on the longest edge, which is more than three times the previous capacity. Essa improvement opens up space for detailed analysis of dense screenshots, technical diagrams and visual documents.
In tests, the model extracts information from complex interfaces with greater accuracy. Agentes computer browsing screens directly benefit from this capability. The higher resolution also helps you create higher-quality slides, documents, and designs.
- Suporte to images with resolution three times higher than the previous version
- Análise improved detailed diagrams and screenshots
- Melhoria in tasks that combine vision with agentic reasoning
- Redução errors in complex visual data extraction
- Aplicação practice in workflow automation and software engineering
Avanços in coding and agentic tasks
Opus 4.7 shows impressive gains in software engineering benchmarks. On SWE-bench Verified, it achieves 87.6%, while on SWE-bench Pro it records 64.3%. Esses numbers surpass the performance of Opus 4.6 and put the model ahead of competitors like GPT-5.4 and Gemini 3.1 Pro in several categories.
Desenvolvedores can delegate difficult tasks with more confidence. The model follows instructions more literally and maintains rigor in operations that span many steps. Ele also introduces a new effort level called xhigh, which is between high and max.
Esse tuning allows finer control between depth of reasoning and response time. No Claude Code, the xhigh level comes as standard. Anthropic recommends starting with high or xhigh in coding tests and agentic flows.
Como the model improves the daily lives of users
Profissionais who use AI for work notice a difference in the first response. Opus 4.7 reduces the need to repeatedly refine prompts. Ele adapts the tone according to the context, whether in corporate emails or technical explanations.
In brainstorms, the model proposes less predictable angles. Ele explains complex subjects by adjusting the level of detail to the target audience. Tarefas, like executive summaries, document analyzes and roadmaps, gains in coherence and adequacy.
Understanding context in long conversations has also evolved. The model tracks information provided several exchanges ago without losing the thread. Isso benefits extensive programming or project planning sessions.
Mudanças affecting project migration
Quem migrates from Opus 4.6 needs to pay attention to some points. The new tokenizer processes text differently and can increase token consumption by up to 35% depending on the content, especially code.
Três API parameters have been removed: temperature, top_p and top_k. Qualquer code that depended on them now returns an error. thinking.budget_tokens has also left the scene and has been replaced by effort levels.
Raciocínio traces appear hidden by default. Usuários who need to see the internal process must adjust settings. Apesar of these changes, Anthropic positions the upgrade as straightforward for most cases.
Limitações and points of attention when using
Factual Erros can still happen, as in any language model. Human review remains essential in critical areas such as legal, medical or journalistic content. Opus 4.7 works as a support tool and does not replace experts.
In some specific benchmarks, such as long context recovery, the model showed mixed results in initial user tests. Anthropic explains that it prioritizes applied capacity over artificial tests with distractors.
Access to Brasil follows the global model. Common Usuários depend on Pro, Team, or Enterprise plans in claude.ai. Desenvolvedores use the API directly or through partner cloud platforms.
Disponibilidade and integration options
Opus 4.7 has been generally released since April 16th. Ele appears in all Claude products and integrations with cloud providers.
Preços remain the same: $5 per million input tokens and $25 per million output tokens. Não there is extra charge for long context.
Empresas who already use Anthropic via Amazon Bedrock, Google Vertex AI or Microsoft Foundry receive the update automatically in supported regions. GitHub Copilot also offers the model for Pro+, Business, and Enterprise users.
Ficha summary technique of Opus 4.7
- Modelo main of the Claude family for professional and agentic tasks
- 1 million token context Janela at no additional cost
- Processamento of images up to 2576 pixels on the long edge
- Novo xhigh effort level for fine control of reasoning
- Desempenho top in SWE-bench Pro (64.3%) and Verified (87.6%)
- Disponível on claude.ai, API and cloud platforms
- Preço identical to Opus 4.6
- Foco on consistency, self-checking and literal instructions
Anthropic continues to release updates in short cycles. Opus 4.7 reinforces the company’s path toward models useful in real work, with safeguards that limit high-risk uses.
Quem already integrates AI into development or professional analysis routines and tends to notice gains in autonomy from the first sessions. The model delivers responses with less manual intervention and greater alignment with what was requested.
