Claude Opus 5 nears Fable at half price, Anthropic says

Anthropic on Friday launched Claude Opus 5, saying the model gets close to Claude Fable 5 on coding and knowledge-work tasks while costing about half as much. The pitch is aimed less at a leaderboard win than at a practical problem for business users: what a stronger model costs once it is used every day.

In its launch note, Anthropic said Opus 5 will cost $US5 (about $7.70) per million input tokens and $US25 (about $38) per million output tokens, matching Opus 4.8. Reuters described the release as an efficiency upgrade, with Anthropic trying to widen access to higher-end performance without shifting customers into the Fable 5 price band.

That price-performance gap is the commercial change. Rather than pushing customers straight to its most expensive tier for stronger coding and reasoning, Anthropic is asking developers and enterprises to treat Opus 5 as enough for routine engineering, documentation and internal workflow jobs. Those workloads can look modest in pilot form, then become expensive when hundreds of staff start using them across a normal working week.

For Australian software teams, the cost question may matter more than another benchmark chart. Code assistants, internal search, support workflows and document-heavy knowledge tasks all scale with use. Once a service is deployed across a larger team, token pricing and tool overhead can weigh as heavily as raw model quality in procurement decisions. Finance and engineering leads will read the claim differently, but both will be looking for evidence in their own logs rather than Anthropic’s charts.

ZDNET said Anthropic is aiming the release at developers and enterprises, citing stronger coding performance, better reasoning efficiency and tool changes designed to work with prompt caching across repeated sessions. The publication said Anthropic claims Opus 5 comes within 0.5 per cent of Fable 5 on CursorBench at max effort, narrowing the gap between its premium model and a cheaper tier likely to handle daily engineering work.

In Anthropic’s announcement, Cursor co-founder Sualeh Asif described the model as a practical upgrade for coding teams.

“Claude Opus 5 delivers near Fable 5 intelligence at Opus speed and cost.”

Sualeh Asif, quoted by Anthropic

The quote is part of Anthropic’s sales material, so buyers will still want their own tests. It does, however, show the audience Anthropic wants: teams that need a stronger coding and reasoning model without taking on the full cost of the company’s most capable tier. That also explains why the launch leans on software workflow examples instead of presenting Opus 5 as a pure benchmark event.

Anthropic product leader Dianne Penn put the same point in cost-and-access terms in comments to Reuters.

“We’re building and continue to consistently deliver frontier intelligence and bring that as accessibly as possible with every model generation.”

Dianne Penn, Reuters

For local buyers, the pricing signal is the story. Anthropic is arguing that AI competition is shifting from the strongest frontier result to models that get close enough on daily coding and knowledge work at a lower operating cost. Australian enterprises, startups and software teams will now have to test that claim against rival offers from OpenAI and Google.