Anthropic released Claude Opus 5 yesterday, and the pitch is unusual for a flagship model launch: it’s not the company’s smartest model, but it’s supposed to get close at roughly half the price of the model it’s chasing. Opus 5 is now the default model on Claude Max and the strongest option available on Claude Pro, and it’s live across every platform today — no waitlist, no staged rollout.
What’s New in the Claude Opus 5 Launch
Anthropic is positioning Opus 5 as approaching the frontier intelligence of Claude Fable 5, the company’s top-tier model, while costing about half as much to run. That’s a different pitch than “our best model got better” — it’s “here’s a model that gets you most of the way to our best model without the bill.” On coding and knowledge-work evaluations, including Frontier-Bench and GDPval-AA, Opus 5 sets a new high-water mark among Anthropic’s own lineup, though the company is upfront that it still trails on some cybersecurity-specific tasks. Opus 5 also ships with a Fast mode that runs about 2.5x the default speed for roughly double the base price — useful if you’re paying for latency, not just capability.
Pricing and Where to Get It
Anthropic held pricing flat at $5 per million input tokens and $25 per million output tokens — the same rate as the outgoing Opus 4.8. That matters more than it sounds: normally a “smarter” model launch comes with a price bump, not a hold. Opus 5 is available now on Claude.ai, the API, and Claude Code, and it’s the model Claude Max subscribers get by default starting today.
The Safety Numbers Anthropic Is Leading With
Anthropic is also using this launch to talk about alignment more than usual. The company’s automated behavioral audit scored Opus 5 at 2.3 on overall misaligned behavior — the lowest (best) score of any recent Anthropic model — with the lowest rate of deceptive behavior and the least susceptibility to being tricked into misuse among the models tested. Whether that number means much to an average user is debatable, but it’s a deliberate contrast to OpenAI’s GPT-5.6, which is currently delayed pending a government oversight review — Anthropic is clearly trying to make “safety-tested” part of what you’re paying for, not just an afterthought.
Why Anthropic Is Moving This Fast
Opus 5 is Anthropic’s fourth model release in two months, which is an aggressive cadence even by 2026’s AI standards. Some of that speed is competitive pressure: Google is already training Gemini 4 internally while Gemini 3.5 Pro still hasn’t had a full public release, and Anthropic clearly doesn’t want to cede the “most capable model you can actually use today” title for long. It also lines up with how much compute Anthropic has been locking down lately — the same company reportedly negotiating a custom AI chip with Samsung ahead of a possible IPO needs a steady drumbeat of releases to justify that spend to investors and enterprise customers alike.
What This Means for You
If you’re already a Claude Max subscriber, you don’t need to do anything — Opus 5 is your default model as of today. Pro users get access to Opus 5 as their strongest available option, and API users can switch by specifying the new model ID at the same rate they were already paying for Opus 4.8, so there’s no real reason to stay on the old version unless you have a workflow specifically tuned to Opus 4.8’s quirks. For developers, the flat pricing is the part worth paying attention to: getting a meaningfully stronger model at the same per-token cost is rare enough in this market that it’s worth re-testing any pipeline you built around the previous generation before assuming nothing changed. If you’re weighing whether to trust an AI model with more autonomous tasks in the first place, that’s a separate question worth thinking through — we cover the tradeoffs in our guide to AI browser agents.
Sources: Axios, TechCrunch, The Next Web, Anthropic.
Deixe um comentário