Anthropic launches Claude Opus 5, nearly matching Fable 5 at half the price
- Claude Opus 5 launched July 24, 2026, priced the same as predecessor Opus 4.8 but delivering much higher performance, while staying close to but behind flagship Claude Fable 5 in overall capability.
- On Frontier-Bench v0.1, Opus 5 surpasses every other model and more than doubles Opus 4.8's score at a lower cost per task; on CursorBench 3.2 at max effort it lands within 0.5% of Fable 5's peak score but at half the cost per task.
- On ARC-AGI 3, Opus 5 scores three times higher than the next-best model; on OSWorld 2.0 it beats Fable 5's best result at just over a third of the cost; on Zapier AutomationBench its pass rate is about 1.5x the next-best model at equal cost.
- Opus 5 gains 10.2 percentage points over Opus 4.8 on an internal organic chemistry benchmark (spectroscopy-based molecular structure inference) and 7.7 points on protein sequence-function prediction tasks.
- Opus 5's safety classifiers now allow source-code vulnerability discovery at all access levels (unlike Fable 5) while still blocking vulnerability discovery in compiled binaries, a change Anthropic frames as supporting defensive cybersecurity work.
Hacker News 의견들
Cost is really the whole story here, right? They're pushing hard on price for performance this time.
Wait, if Opus 5 isn't more capable than Fable 5, what's even the point of releasing it?
Presumably it's cheaper, that's the point. Fable is twice the price so Opus gives you most of the capability for half the cost.
It's better than Opus 4.8, that's the point people are missing. Nobody expects a Sonnet release to beat Opus either.
Kind of feels like the illusion of progress to appease shareholders and postpone the AI bubble popping, honestly.
I use Opus for coding delegation and Fable for planning, so having a cheaper strong Opus is genuinely useful to me.
Their own benchmarks show Opus 5 spanking Fable 5 pretty soundly in most tests, so the framing in the blog post is confusing.
They highlighted Opus 5 as best for agentic coding even though the number is slightly lower than Fable. Close enough for marketing I guess.
So almost Fable 5 performance at half the cost? Sign me up, that's a great deal.
This makes Sonnet 5 look even worse. There's no reason to use Sonnet over Opus now with low or no reasoning.
My understanding is Opus is for planning and architecture, Sonnet for actual execution, cheaper that way with minimal quality loss.
New model is nice but they should fix their existing infra first. We're considering dropping our Claude Team sub, it's been unusable lately with dropped sessions and http errors.
Why can't Fable get the same source-code vulnerability discovery permission as Opus 5? If the safeguards are supposedly identical except for this one change, seems arbitrary.
Because their model got flagged by the government before for this kind of thing and they don't want a repeat.
My drug discovery malaria screening project got blocked by Fable over bio-security concerns. Guess that domain stays unsupported by Anthropic.
Noticed none of the comparisons mention Kimi K3. Is there a comparison chart anywhere?
That's by design, Anthropic wants open-weight models regulated hard, so they're not going to give Kimi K3 free attention.
I'm honestly confused what the real difference between Opus 5 and Fable 5 is now beyond cost and marginal capability.