Anthropic dropped a new model on Thursday. It’s called Claude Opus 5.
The timing is… deliberate. Weeks after a bruising standoff with the US government over a different model, and just days after OpenAI dealt with a security mess, Anthropic went live.
They say Opus 5 “comes close to the capabilities” of the controversial Fable 5 model in many areas. Fable 5, if you remember, is the Mythos-class model that got pulled offline. It sparked a regulatory firestorm with the Trump administration. Now it’s back, locked down tighter than a drum with new cyber safeguards. Opus 5 borrows that power. Specifically for complex coding.
“We continue to work with our government partners… This includes Opus 5,” said spokesperson Danielle Ghiglieri.
They didn’t give a hard “yes” to whether the Trump administration blessed Opus 5 before launch. They said they’re continuing to cooperate. Independent testing is happening. That’s the vibe.
Which Claude model fits enterprise knowledge work?
Anthropic isn’t trying to replace Fable 5 entirely. Not yet.
Opus 5 is pitched as the “ideal everyday model.” It’s for enterprise. For knowledge workers. For biology. If you’re running long-horizon AI agents or pushing the absolute limits of ambition? Stick with Fable 5.
But Opus 5 has something else going for it. It’s safer.
The company calls it “the most aligned Opus model.” Less susceptible to tricks. Less prone to misuse. Compared to Opus 4.8? Stronger cybersecurity safeguards. In an era where AI regulation is no longer a hypothetical, alignment is marketing currency.
How does Opus 5 pricing compare to GPT-5.6?
There’s a lot of noise right now about AI costs. Lawsuits over hidden caps. Complaints about labs passing expenses onto users. Anthropic addressed this head-on.
Prices didn’t go up.
Opus 5 costs the same as its predecessor, Opus 4.8. That’s $5 per million input tokens. $25 for output.
Compared to the heavy hitter? It’s half the price of Fable 5.
Compared to the competition? It’s slightly cheaper than OpenAI’s GPT-5.6.
You can pay for speed, though. “Fast mode” is launching today in research preview. Higher throughput. Double the price. It’s a trade-off. Speed costs money.
There’s another tweak for enterprise users. You can opt-in to automatic fallbacks. If Opus 5’s safeguards trigger a decline—say, for safety reasons—it automatically routes the request to a lower-tier model. No manual intervention needed. Just a smooth(ish) degradation of service.
Is this the new normal? Government oversight, price stability, and safety fallbacks? Probably.
The market is settling. Opus 5 sits in the middle. Powerful enough for coding. Safe enough for the feds. Cheap enough for the wallet.
It’s a calculated move.
