Anthropic released Claude Opus 4.7 on Thursday, pitching it as a direct upgrade to Opus 4.6 for software engineering and long-running agent work. The announcement post emphasizes gains on the hardest coding tasks, where the company says users can hand off work that previously needed close supervision.
Pricing holds at $5 per million input tokens and $25 per million output, the same as Opus 4.6. But there's a catch: an updated tokenizer can map the same text to 1.0 to 1.35 times more tokens. So effective cost creeps up.
The interesting stuff sits around autonomy. Opus 4.7 adds a new xhigh effort level between high and max, and Claude Code now defaults to it. Vision also got a real bump, up to 2,576 pixels on the long edge, roughly three times the prior ceiling.
Early testers say the numbers match the pitch, though they're all customers. Cursor CEO Michael Truell called it "a meaningful jump" on CursorBench, clearing 70% versus 58% on Opus 4.6. XBOW reports 98.5% on its visual-acuity benchmark against 54.5% for the prior model. These are self-reported by partners; independent verification hasn't landed.
Opus 4.7 sits below Mythos Preview, Anthropic's more capable model kept on a tight leash over cybersecurity concerns flagged under Project Glasswing. Anthropic says 4.7 ships with safeguards that block prohibited cyber uses, and legitimate security researchers can apply to a verification program for access. Alignment specifics are in the system card.
Bottom Line
Claude Opus 4.7 ships today at the same $5/$25 per million tokens as Opus 4.6, with a new xhigh effort tier and triple the image resolution.
Quick Facts
- Release date: April 16, 2026
- Pricing: $5 per million input tokens, $25 per million output (unchanged)
- Max image input: 2,576 pixels on long edge, roughly 3x prior Claude models
- New xhigh effort level between high and max
- API model string: claude-opus-4-7
- Benchmark gains are company-reported or cited by early-access partners
