Anthropic Launches Claude Opus 4.7 With Enhanced Coding and Cyber Safeguards

Alvin Lang Apr 16, 2026 14:55

Anthropic releases Claude Opus 4.7 with 13% coding benchmark gains, new cybersecurity restrictions, and improved vision capabilities at unchanged API pricing.

Anthropic Launches Claude Opus 4.7 With Enhanced Coding and Cyber Safeguards

Anthropic just dropped Claude Opus 4.7, and the $380 billion AI company isn't messing around. The new model delivers double-digit improvements in coding benchmarks while introducing automated safeguards that block high-risk cybersecurity requests—a direct response to concerns that forced the company to keep its more powerful Mythos Preview model under wraps.

The release comes weeks after Anthropic announced its Claude Mythos Preview was too dangerous for public deployment, with sources indicating government officials were briefed on its advanced cyber capabilities.

What's Actually Different

Opus 4.7 scored 72.5% on SWE-bench Verified at high effort versus 68.6% for its predecessor. Early testers reported more meaningful gains on real-world tasks. Cursor's internal benchmark showed a jump from 58% to 70%. Rakuten claims the model resolves 3x more production tasks than Opus 4.6.

The vision upgrade might matter more than the headline numbers suggest. The model now processes images up to 2,576 pixels on the long edge—roughly 3.75 megapixels, or three times previous limits. XBOW's penetration testing team reported visual acuity scores jumping from 54.5% to 98.5%, effectively unlocking an entire class of computer-use applications.

Memory handling got a quiet but significant boost. The model now maintains context across multi-session work using file system-based memory, reducing the context-stuffing that drives up token costs.

The Cyber Angle

Here's where it gets interesting for anyone tracking AI risk. Anthropic explicitly stated they "experimented with efforts to differentially reduce" Opus 4.7's cyber capabilities during training. The model ships with automated detection systems that block prohibited cybersecurity uses.

Security professionals wanting legitimate access—vulnerability research, penetration testing, red-teaming—can apply through a new Cyber Verification Program. This two-tier approach signals Anthropic's strategy for eventually releasing Mythos-class models: prove the safeguards work on less capable systems first.

Pricing and Access

API pricing stays flat at $5 per million input tokens and $25 per million output tokens. The model is live across Claude's consumer products, Amazon Bedrock, Google Cloud's Vertex AI, and Microsoft Foundry. Developers can hit it via claude-opus-4-7.

One catch for existing users: the updated tokenizer means identical inputs can map to 1.0–1.35x more tokens depending on content type. Anthropic's migration guide recommends measuring real traffic impact before assuming costs stay constant.

What Traders Should Watch

Anthropic's February Series G valued the company at $380 billion—roughly on par with Nvidia's market cap when it first cracked the AI narrative. The Opus 4.7 release, combined with the company's April expansion of its Google and Broadcom compute partnership for "multiple gigawatts" of next-gen infrastructure, suggests the private company is positioning for either an IPO or strategic acquisition.

For public market exposure, watch enterprise software companies citing Claude integration in earnings calls. Harvey, Notion, Replit, and Vercel all provided testimonials for this release—their deployment patterns could signal broader AI infrastructure spend trends heading into Q3.

Image source: Shutterstock