60 percent. That is the approximate price drop between the top-tier Fable and the new Opus 5. For those of us tracking the token wars, that isn’t just a discount; it’s a signal that the era of the “prestige model” is ending.
Anthropic just dropped Opus 5, and as TechCrunch reports, the new model is both cheaper and significantly less restrictive than its predecessor, Fable. For a long time, using Anthropic’s top-end models felt like chatting with a very polite, very nervous librarian who might stop mid-sentence to remind you that it is important to be inclusive when discussing the history of salt mining. Fable was the peak of this “Constitutional AI” philosophy—hyper-safe, incredibly expensive, and often so cautious it became useless for complex, edge-case coding tasks.
Opus 5 changes the math. By lowering the cost and loosening the safety filters, Anthropic is finally admitting that developers don’t want a moral compass; they want a tool that doesn’t lecture them. (Assuming the API doesn’t melt under the sudden migration from Fable). It is a pragmatic pivot. They have spent years building a brand around safety, but safety doesn’t win market share when the competition is offering raw power and permissive prompts.
Do we actually care about the “Constitution” if it comes at the cost of a 40% increase in refusals? Probably not.
The real story here isn’t the price point, but the implicit admission that Fable was a strategic misstep. Fable was positioned as the ultimate intelligence peak, but it suffered from the same problem as luxury cars that are too precious to actually drive. It was a trophy model. Now, with Opus 5, Anthropic is essentially releasing a stripped-down, high-performance version of their tech to stop the bleed of users moving toward open-weight alternatives.
It is like a high-end fashion house suddenly releasing a line of durable workwear because they realized nobody can actually live in a silk gown. They are trading the aura of “safest model in the world” for actual utility. The restriction drop is the most interesting part. If Opus 5 is indeed “less restrictive,” it means the internal alignment tax has been lowered. This is where the real performance gains usually hide. When a model stops spending half its compute wondering if its answer might offend a hypothetical person, it actually has more headroom to solve the logic puzzle you put in front of it.
But this creates a weird identity crisis for the lab. You cannot claim to be the “Safety First” company while simultaneously undercutting your own safety-heavy model with a cheaper, looser alternative. Either the safety filters in Fable were overkill, or the filters in Opus 5 are insufficient. There is no middle ground where both are “correct.”
Fable is now a ghost.
The market will move fast on this. There is no reason for a mid-sized dev shop to keep paying the Fable premium when Opus 5 does the heavy lifting for a fraction of the cost. By Q4, we will see Fable relegated to a niche enterprise tier, sold exclusively to the most risk-averse banks and government agencies who are more afraid of a PR scandal than they are of high latency.
The move is a surrender to the reality of the current market. Developers are tired of the hand-holding. We want models that can handle the grit of real-world data without triggering a lecture on AI ethics every three prompts. Anthropic finally got the memo. Whether they can maintain their performance lead while slashing prices and filters remains to be seen, but at least they are finally playing the same game as everyone else.
