Default to Opus 5 for most knowledge work

Opus 5 is half the price with better benchmarks on agentic knowledge work. Reserve Fable 5 for long-horizon autonomous runs


Opus 5 is the new highest performing AI model on the Artificial Analysis index, breaking a performance plateau which had held for 45 days, displacing Anthropic’s own twice-as-pricey Fable 5.

On the AA Intelligence Index it’s 61 vs 60 – effectively tied. The gap is elsewhere. Opus 5 beats Fable 5 on agentic knowledge-work benchmarks (by +114 Elo on GDPval-AA v2 and +146 on AA-Briefcase). So Opus 5 is better at exactly the professional-output work most knowledge workers will be concerned about, at half the token price and 26% lower cost per task.

Why the price difference?

  • Fable 5 is a substantially larger mode so Anthropic can’t price it below the inference economics that demands – regardless of of how Opus 5 peforms on benchmarks.
  • Fable 5 still leads on factual knowledge (AA-Omniscience) and hallucinates less — Opus 5’s hallucination rate rose 14 points to 50% when uncertain. For recall-heavy or citation-sensitive work it remains the better bet.
  • Anthropic’s own claim is that Fable’s lead grows with task length and complexity. Index tasks are relatively bounded, so the benchmark may not capture the full extent of the benefits Fable 5 is offering.
  • Pricing the way they have will drive more buyers to Opus 5 for more tasks. Given the lower inference costs, Opus is the model they would prefer you to buy.

Right now you are best defaulting to Opus 5 for most knowledge work tasks, reserving Fable 5 for long-horizon autonomous runs and recall-critical tasks. And of course, expect un update on Fable before any of us can get comfortable.

That is, of course, if you haven’t already bailed for an open weights competitor at an even lower price point.

The good news all round is that the price/performance of AI continues to tumble. Google has been moving too with 1M token context window models for as little as $1.16 $/M blended.

Worth noting one other impact on our tracker: AA rescored several mid-tiers this week and Kimi K2.6 fell from 44 to 35, Mercury 2 from 25 to 21 and Grok 4.3 from 38, to 36, due to a change in methodology.

Default to Opus 5 for most knowledge work

More HFS Quick Takes

Sign In

Sign up for a free
research account

With the exception of our Horizons reports, most of our research is available for free on our website. Sign up for a free account and start realizing the power of insights now.

By registering you agree to our privacy policy.

I hereby consent that HFS Research can process my personal data.

Digests/Newsletters: Overviews of the latest news, insight, and research by HFS.

HFS Events: Exclusive invitations to HFS webinars, roundtables, and summits, bringing together key industry stakeholders focused on major innovations impacting business operations.

Premium Access

Our premium subscription gives enterprise clients access to our complete library of proprietary research, direct access to our industry analysts, and other benefits.

Contact us at [email protected] for more information on premium access.

    Contact Ask HFS AI Support