On April 16, 2026, Anthropic released Claude Opus 4.7 — the most capable version of its AI assistant that anyone can use today. It writes code more carefully, sees images in higher resolution, and is designed to work on long, complicated tasks without constant hand-holding. Here’s what that actually means, who Anthropic is, and where Opus 4.7 fits in the wider AI landscape.
First, Who Is Anthropic?
Anthropic is an AI company founded in 2021 by former OpenAI researchers, including siblings Dario and Daniela Amodei. From the start, the company has positioned itself as the “safety-first” lab — the one that wants to build powerful AI but also wants to understand exactly what that AI is doing and why. That’s not just branding: Anthropic publishes unusually detailed research on how its models behave, where they fail, and what risks they might pose.
The company’s main product is Claude, a family of AI assistants available through a web app (claude.ai), a mobile app, a developer API, and integrations with cloud platforms like Amazon Bedrock, Google Vertex AI, and Microsoft Foundry.
The Claude Model Family
Anthropic structures its Claude models into three tiers, each optimised for a different tradeoff:
- Haiku — the fast, lightweight model. Good for quick tasks, classification, high-volume workloads where speed and cost matter more than raw intelligence.
- Sonnet — the balanced model. The everyday workhorse for most users: smart enough for real work, fast enough for real conversations.
- Opus — the most capable model. The one you reach for when the problem is genuinely hard: complex code, long documents, deep research, or multi-step tasks that need careful reasoning.
Claude Opus 4.7 sits at the top of this stack — with one asterisk, which we’ll get to shortly.
What’s New in Opus 4.7
Opus 4.7 is a targeted upgrade to Opus 4.6, which launched in February 2026. It’s not a generational leap — think of it as a sharpening of the blade rather than a new weapon. The gains are concentrated in the areas that matter most to the people using AI for serious work.
Better at coding — and at checking its own work
This is the headline. Opus 4.7 is noticeably better at handling complex, long-running software engineering tasks, and — crucially — it’s designed to verify its own outputs before reporting back. Early testers describe being able to hand off hard coding problems that previously required close supervision and actually trust the result.
High-resolution vision
Opus 4.7 is the first Claude model with high-resolution image support, accepting images up to roughly 3.75 megapixels — more than triple the previous limit. That matters for anything where fine visual detail counts: reading diagrams, parsing dense documents, driving computer-use agents that need to interact with real interfaces.
Finer control over reasoning depth
Developers can now choose from effort levels — including a new “xhigh” setting that sits between “high” and “max.” In plain English: you can tell Claude how hard to think about a problem, trading off speed and cost against depth of reasoning. For tough coding work, Anthropic recommends starting at xhigh.
Task budgets
A new API feature lets developers set a rough token budget for a whole agentic task — thinking, tool calls, results, and final output all included. This gives teams running long autonomous workflows a practical ceiling on how expensive a single run can get.
Better memory across sessions
Opus 4.7 is also better at using file-system-based memory — the kind of persistent notes that agents leave for themselves to pick up work later. That makes it more useful for genuinely long-horizon projects where context builds up over days or weeks.
How Does It Compare to OpenAI and Google?
On Anthropic’s own benchmarks, Opus 4.7 beats OpenAI’s GPT-5.4 (released in early March 2026) and Google’s Gemini 3.1 Pro on a number of key measures — agentic coding, scaled tool use, agentic computer use, financial analysis. But the race is tight: on directly comparable benchmarks, Opus 4.7 leads GPT-5.4 by a margin of roughly 7-to-4, and competitors still hold leads in specific areas like agentic search and multilingual Q&A.
The practical takeaway is that no single model is universally best in 2026. Opus 4.7 is positioned as the strongest choice for long-running, autonomous, reliability-critical work — the kind of task where you need the AI to plan carefully, use tools properly, and not go off the rails halfway through.
The Mythos Asterisk
There’s a more powerful Claude model than Opus 4.7 — it’s called Claude Mythos Preview, and Anthropic has deliberately chosen not to release it broadly. Mythos has been made available to a handpicked group of cybersecurity and technology companies under a programme called Project Glasswing, where it’s being used to find and patch security vulnerabilities in enterprise software.
The reason for the restricted release is safety: Mythos is capable enough that Anthropic worries about how it could be misused if deployed too widely. Opus 4.7, by contrast, has had some of its cyber capabilities deliberately reduced during training, and ships with automated safeguards designed to detect and block high-risk cybersecurity requests.
For almost everyone, Opus 4.7 is the most capable Claude you can actually get your hands on.
Where to Use It and What It Costs
Claude Opus 4.7 is available from day one across every Claude surface: the claude.ai web app, the mobile apps, Claude Code (Anthropic’s terminal-based coding tool), the API, and through Amazon Bedrock, Google Vertex AI, and Microsoft Foundry. GitHub Copilot is also rolling it out to Pro+, Business, and Enterprise users.
Pricing is unchanged from Opus 4.6: $5 per million input tokens and $25 per million output tokens. That’s cheap by flagship-model standards, and the fact that Anthropic didn’t raise prices with the upgrade is notable.
Why This Release Matters
Zoom out and Opus 4.7 says something about where the AI industry is heading in 2026. The flashy consumer announcements of 2023 and 2024 have given way to a quieter, more serious phase: models that can actually do work, run for hours without supervision, use tools reliably, and be trusted with real production systems. The improvements in Opus 4.7 — self-verification, xhigh effort, task budgets, better memory — are all infrastructure for that world.
If you’re building with AI, Opus 4.7 is worth trying on your hardest problem. If you’re watching from the sidelines, it’s a useful checkpoint on how far the frontier has moved — and how carefully the companies at that frontier are now choosing what to release and what to hold back.
Leave a Reply