<p>Anthropic has released Claude Opus 4.7, roughly two months after Opus 4.6. The headline improvements are in agentic coding, long-running task handling, and multimodal understanding.</p>
<h2>What’s actually different</h2>
<p>The model carries context across sessions and handles ambiguity better than 4.6. In practice, that means you can hand it a multi-file refactor or a debugging task that spans an entire service without it drifting halfway through. It also verifies its own outputs before reporting back, which cuts down on the “looks done but isn’t” problem.</p>
<p>On benchmarks: 64.3% on SWE-bench Pro, 87.6% on SWE-bench Verified, and 70% on CursorBench (up from 58% on 4.6). Early testers report 10–14% gains on coding tasks and roughly 3x more production task resolution on SWE-bench variants.</p>
<p>Image support got a meaningful upgrade too — the model now handles images up to 2,576px on the long edge (~3.75MP), which makes a real difference on dense charts, UI screenshots, and scanned documents.</p>
<p>A new <code>xhigh</code> effort level sits between <code>high</code> and <code>max</code>. Worth knowing: Anthropic has raised Claude Code’s default effort from <code>medium</code> to <code>xhigh</code>, so token usage will likely go up unless you adjust it manually.</p>
<p>The context window is 1M tokens. Pricing stays at $5/M input and $25/M output, with up to 90% savings via prompt caching.</p>
<p><strong>One migration note:</strong> there’s a new tokenizer that may increase token counts by 1.0–1.35x compared to 4.6.</p>
<h2>New features in Claude Code</h2>
<p>Max users get auto mode, which removes permission prompts so you can hand off a task and come back to finished work. There’s also a new <code>/ultrareview</code> command.</p>
<h2>Where it’s available</h2>
<ul>
<li>Claude Pro, Max, Team, and Enterprise plans</li>
<li>API (model ID: <code>claude-opus-4-7</code>)</li>
<li>Amazon Bedrock (US East, Asia Pacific Tokyo, and two European regions; up to 10,000 RPM)</li>
<li>Google Cloud Vertex AI</li>
<li>Microsoft Foundry</li>
<li>Vercel AI Gateway (<code>anthropic/claude-opus-4.7</code>, no markup, no separate Anthropic account needed)</li>
<li>GitHub Copilot for Pro+, Business, and Enterprise — replaces Opus 4.5 and 4.6 in the model picker, with a 7.5× premium request multiplier at promotional pricing until April 30th</li>
<li>Devin’s agent harness, also at promotional pricing until April 30th</li>
<li>v0 (50% off for two weeks)</li>
</ul>
<p>Teams at Replit, Vercel, Cursor, Notion, Bolt, Augment Code, and Devin have all reported double-digit gains in coding, document reasoning, and agentic pipelines. Opus 4.7 is also now part of MagicPath, where early testing shows stronger image-to-code performance and cleaner React component output.</p>


Collections

Anthropic has released Claude Opus 4.7, featuring improvements in agentic coding, long-running task handling, and multimodal understanding. Key benchmarks include 64.3% on SWE-bench Pro and 87.6% on SWE-bench Verified, with early testers reporting 10–14% gains on coding tasks. Image support now handles up to 2,576px on the long edge. A new `xhigh` effort level has been added, and Claude Code's default effort has been raised from `medium` to `xhigh`, which may increase token usage. The context window is 1M tokens with pricing unchanged at $5/M input and $25/M output. A new tokenizer may increase token counts by 1.0–1.35x. The model is available across Claude plans, API, Amazon Bedrock, Google Cloud Vertex AI, Microsoft Foundry, Vercel AI Gateway, GitHub Copilot, Devin, v0, and more.

Claude Opus 4.7 announced with improved instruction-following and self-verification

<p>Seems like ai tools are getting better and better. It is difficult to keep up with everything</p>


<p>Bro we can’t catch up the updates AI moving so fast</p>
