anthropic.com via Hacker News

Anthropic Ships Opus 5, Pitches Half-Price Rival to Fable 5

TL;DR

  • Anthropic released Claude Opus 5 at $5 per million input tokens and $25 per million output tokens, the same list price as Opus 4.8.
  • The company positions Opus 5 as close to the frontier intelligence of Claude Fable 5 at half the price, backed by curated customer quotes.
  • Anthropic says cyber-safety classifiers trigger about 85% less often on Opus 5 than on Fable 5, with a 2.3 misaligned-behavior score.

Anthropic put a new flagship on the shelf today with an unusually pointed pricing message. In its announcement, the company calls Claude Opus 5 a model that "comes close to the frontier intelligence of Claude Fable 5 at half the price," while keeping the list price at $5 per million input tokens and $25 per million output tokens, the same as Opus 4.8.

The pitch, in other words, is a free capability upgrade for anyone already on Opus 4.8 and a cheaper substitute for Fable 5 at the top tier. Anthropic's own numbers back this in places. It says Opus 5 "more than doubles Opus 4.8's performance at a lower cost per task," and gains 10.2 percentage points over Opus 4.8 on an internal organic-chemistry benchmark. The launch post also lines up curated quotes from Scott Wu at Devin, Sualeh Asif at Cursor, and Wade Foster at Zapier, each landing on the same "near-Fable performance at Opus cost" beat.

The safety framing gets its own moment. Opus 5 scores 2.3 on Anthropic's overall misaligned-behavior audit, described as the lowest of any recent Anthropic model, and Anthropic expects its cyber-safety classifiers to intervene "around 85% less often than they do for Fable 5." Both are Anthropic's own measurements on Anthropic's own audits, so read them as a directional signal rather than an independently replicated result.

The honest caveat is that almost every benchmark cited here is either internal to Anthropic or run by paying customers quoted in the same launch post. Independent evaluators have not published head-to-head numbers yet, and the "half the price of Fable 5" line depends on how Fable 5's own list pricing is being compared. What the reporting does not give you is how Opus 5 stacks up on the hardest cybersecurity and biology tasks, which is where regulated buyers will actually stress-test it.

If the pricing claim survives third-party evaluation, though, the interesting move is not at Anthropic. It is at everyone shipping agents on top. Devin, Cursor, and Zapier just got a cheaper knob to turn on per-task cost, and the frontier tier's price floor moved with them.

Shared on Bluesky by 7 AI experts (top 5 by trust)