April 16, 2026
New brain, same drama
Claude Opus 4.7
Claude Opus 4.7 drops: smarter code, stricter locks, and the Mythos tease continues
TLDR: Anthropic launched Claude Opus 4.7 with better coding, vision, and memory at the same price. Comments split between happy devs and frustrated users accusing “token shrinkage,” plus a safety brawl over new cybersecurity blocks and eye rolls at the never-ending Mythos tease—making this upgrade feel powerful but polarizing.
Claude Opus 4.7 just landed with promises of sharper coding autopilot, better “eyes” for images, stronger memory, and the same price as before—see the announcement. But the comments? Pure fireworks. Early testers rave that they can hand off difficult projects without babysitting, and note it follows instructions to the letter now. Cue a wave of “re-tune your prompts” warnings and a few victory laps from devs who’ve wanted a more literal assistant. Meanwhile, a top-voted gripe from one user: they’re “excited to use 1 prompt and have my whole 5-hour window at 100%,” grumbling about session limits and “token shrinkage and gaslighting.” Ouch.
The biggest brawl is over security. Anthropic added guardrails to block high‑risk hacking requests and is testing them here before unleashing its most powerful model, Mythos. One security pro calls that choice “potentially fatal,” arguing you need equal firepower to fight attackers. Others shrug and point to the new Cyber Verification Program: legit researchers can still get in—if they apply. Then there’s the drama over Mythos: fans keep seeing it in charts, not their accounts. The community’s meme of the day: “carrot on a stick.” One confused commenter summed it up best: “someone tell me if I should be happy.” Verdict from the crowd? Opus 4.7 looks like a real upgrade… wrapped in guardrails, queues, and a Mythos-shaped cliffhanger.
Key Points
- •Claude Opus 4.7 is generally available with improvements in advanced software engineering, strict instruction following, and self‑verification.
- •The model’s vision supports images up to 2,576 pixels on the long edge (~3.75 MP), enabling higher-fidelity multimodal tasks.
- •Opus 4.7 includes cybersecurity safeguards that detect and block prohibited or high‑risk uses; cyber capabilities were reduced relative to Mythos Preview during training.
- •Availability spans Claude products, API, Amazon Bedrock, Google Cloud’s Vertex AI, and Microsoft Foundry; pricing remains $5/M input tokens and $25/M output tokens.
- •Early testing shows state‑of‑the‑art results on Finance Agent and GDPval‑AA, improved memory for multi‑session work, and a safety profile similar to Opus 4.6.