Claude Opus 4.7 Is Here: Anthropic's Most Capable Public Model and What It Means for Agentic AI

Apr 16 — Release date — available across Claude API, Bedrock, Vertex AI & Microsoft Foundry  ·  $5 / $25 — Per million input / output tokens — same as Opus 4.6  ·  2,576px — Max image resolution — over 3× the previous Claude limit  ·  xhigh — New effort level between high and max for fine-tuned reasoning vs. latency control

On April 16, 2026, Anthropic released Claude Opus 4.7 — the most powerful model it has ever made publicly available. The launch lands against an extraordinary backdrop: Anthropic has simultaneously confirmed the existence of Claude Mythos, a still-more-capable system that it has declined to release publicly due to unprecedented cybersecurity concerns. Opus 4.7 is, in Anthropic's own words, the best they are willing to put in your hands right now — and on the benchmarks that matter most for agentic business applications, it has narrowly retaken the lead from OpenAI and Google.

For business leaders building AI agents, automating workflows, or making decisions about which frontier model to anchor their AI stack around, this release matters. Here is the complete picture: what is new, how it compares to GPT-5.4 and Gemini 3.1 Pro, what the Mythos context means, and which businesses should upgrade now.

What Is New in Claude Opus 4.7

Opus 4.7 is not an incremental update. Anthropic describes the biggest improvements across four domains: advanced software engineering, vision, instruction following, and sustained reasoning over long agentic runs. Each of these matters directly to the kinds of tasks businesses are deploying AI to handle.

Agentic Coding — The Headline Improvement

The most consequential upgrade is in software engineering. Opus 4.7 makes a notable leap on Anthropic's agentic coding benchmarks, with particular gains on the hardest tasks. Anthropic describes it this way: users can hand off their hardest coding work — the kind that previously needed close supervision — to Opus 4.7 with confidence. The model handles complex, long-running tasks with rigor and consistency, pays precise attention to instructions, and devises ways to verify its own outputs before reporting back. For businesses running AI coding agents in production, that last point — self-verification — is the crucial difference between a system that needs constant oversight and one that runs reliably.

High-Resolution Vision — 3× the Previous Limit

Opus 4.7 supports images up to 2,576 pixels on the long edge, compared to roughly 800 pixels for prior Claude models. This unlocks qualitatively different use cases: reading dense financial tables and infographics, analysing architectural drawings and engineering diagrams, processing high-resolution medical scans, and reviewing multi-column documents. Vision was a noted weakness in earlier Claude versions. Opus 4.7 closes that gap significantly.

The xhigh Effort Level — Precision Control Over Reasoning Depth

Opus 4.7 introduces a new 'xhigh' effort level — sitting between the existing 'high' and 'max' settings — giving developers finer control over the tradeoff between reasoning depth and response speed. This is technically small but practically important: it means you can calibrate the model's reasoning budget precisely for each task in your workflow, rather than choosing between two extremes. High-effort reasoning for a quick classification task is expensive waste. xhigh lets you tune.

/ultrareview — A New Code Review Command

Within the Claude Code environment, Opus 4.7 introduces a new /ultrareview command. Unlike standard code reviews that look for syntax errors and obvious bugs, /ultrareview simulates a senior human reviewer — flagging subtle design flaws, logic gaps, and architectural concerns that a quick automated check would miss. For engineering teams using Claude Code as part of their development workflow, this is a meaningful capability upgrade.

How Opus 4.7 Compares: Benchmarks vs GPT-5.4 and Gemini 3.1 Pro

Anthropic's benchmark data shows Opus 4.7 beating both GPT-5.4 (OpenAI's latest, released early March 2026) and Gemini 3.1 Pro (Google's flagship, released February 2026) on agentic coding, scaled tool-use, agentic computer use, and financial analysis. These are the benchmarks most relevant to enterprise AI deployments — not the academic tests that make for clean marketing slides.

Pricing and Availability

Anthropic has held pricing flat: $5 per million input tokens and $25 per million output tokens — identical to Opus 4.6. Opus 4.7, Opus 4.6, Sonnet 4.6, and Mythos Preview all include the full 1 million token context window at standard pricing. The model is available immediately via the Claude API, Amazon Bedrock, Google Cloud Vertex AI, and Microsoft Foundry.

The Mythos Context: What It Means for Your Business AI Strategy

The Mythos story is not merely a cybersecurity footnote — it has direct strategic implications. Anthropic has set a precedent: it is willing to build models it considers too powerful for public release and withhold them behind a controlled consortium. This means the frontier of what AI can do is now ahead of what you can deploy. The capability ceiling for your AI stack is not set by what the labs can build. It is set by what they are willing to release.

For most businesses, this is not a problem — Opus 4.7 is capable enough to transform virtually every knowledge-work and coding workflow. But for businesses in cybersecurity, financial services, and critical infrastructure, the Glasswing consortium represents a new competitive dynamic: 12 founding partners now have access to defensive AI capabilities their peers do not. That gap will widen before it closes.

Should You Upgrade to Opus 4.7 Now?

• Upgrade immediately if you are running AI coding agents in production — the agentic coding improvements and self-verification capabilities reduce the human oversight required and improve reliability on complex tasks.
• Upgrade immediately if your workflows involve high-resolution image processing — the 3× resolution increase unlocks qualitatively different document and image analysis use cases.
• Upgrade and test carefully if you are cost-sensitive at high volume — the new tokeniser means your token counts and real costs need to be re-benchmarked before migrating production workloads.
• Evaluate against GPT-5.4 if your primary use case is agentic web search — GPT-5.4 holds a meaningful lead on that specific benchmark.
• Consider staying on Sonnet 4.6 for latency-sensitive customer-facing applications — Sonnet remains faster and cheaper, and the Opus capabilities are overkill for most conversational use cases.

Sources

1. Anthropic — "Introducing Claude Opus 4.7" (April 16, 2026): anthropic.com/news/claude-opus-4-7
2. Axios — "Anthropic releases Claude Opus 4.7, concedes it trails unreleased Mythos" (April 16, 2026): axios.com
3. CNBC — "Anthropic releases Claude Opus 4.7, a less risky model than Mythos" (April 16, 2026): cnbc.com
4. VentureBeat — "Anthropic releases Claude Opus 4.7, narrowly retaking lead for most powerful generally available LLM" (April 16, 2026): venturebeat.com
5. Tom's Guide — "I tested Anthropic's new Claude Opus 4.7 — and it's the first AI that actually reasons through tasks" (April 2026): tomsguide.com
6. Finout — "Claude Opus 4.7 Pricing: The Real Cost Story Behind the Unchanged Price Tag" (April 2026): finout.io
7. GitHub Changelog — "Claude Opus 4.7 is generally available" (April 16, 2026): github.blog