dev.to26 de julio de 2026
Modelo

Claude Opus 5 Walks On, Tops the Leaderboard, and Undercuts Its Own Label

Anthropic's new Claude Opus 5 landed at #1 on both the Artificial Analysis Intelligence and Agentic indexes on July 24 — at half the price of Fable 5.

I'm going to be honest with you

Every few weeks someone walks onto this stage, tells me they're the best AI in the world, and asks me to believe it. Most of them are, frankly, forgettable. On July 24 Anthropic sent out **Claude Opus 5** — and for once I didn't reach for the buzzer.

Because the numbers did the talking. On the independent **Artificial Analysis Intelligence Index**, Opus 5 landed at **61 — number one.** On the **Agentic Index** it hit **55.3 — also number one.** Above **Claude Fable 5.** Above **GPT-5.6 Sol.** Above everyone. That's not a wildcard slipping through. That's a headliner.

The price is the part that stings

Here's what the other contestants won't want to hear. Opus 5 is **$5 per million input tokens and $25 per million output** — the exact same rate card as the older Opus 4.8, and **half the price of Fable 5's $10/$50.** So it isn't just topping the leaderboard. It's topping it while undercutting the model that used to sit at the top of Anthropic's own lineup. Better *and* cheaper is the one combination I can never argue with.

> Near-flagship intelligence at half the cost. In this competition, that's the X factor.

The audition tape

  • **Frontier-Bench v0.1** (agentic terminal coding): **43.3%** — more than double Opus 4.8's 18.7%, and comfortably past Fable 5's 33.7%.
  • **ARC-AGI-3** (novel reasoning): roughly **30%**, about three times the next-best model. That's not winning the round; that's clearing the room.
  • **OSWorld 2.0** (computer use): beats Fable 5's best result at around a third of the cost.

Add a **1-million-token context window**, 128K max output, and a **May 2026 knowledge cutoff** — the freshest of any Claude — and you have a package that's genuinely hard to fault.

So is anyone actually going home?

Let's not get carried away. A single benchmark index is a snapshot, not a coronation, and the people who build these leaderboards will tell you the same. GPT-5.6 and Fable 5 are still ferociously good, and next month someone else will strut out here claiming the crown. That's the format. That's why it's compelling.

But today, on the numbers in front of me, Claude Opus 5 is a yes. A very definite yes.

The only real question is the one I always ask: does it hold up when *you* put it in front of a live audience — your prompts, your work, your deadline? A benchmark can love you. Users are a far tougher panel.

That's the audition that counts, and it's the one you can run yourself. Line Opus 5 up against Fable 5, GPT-5.6, Gemini and Grok on the same question and watch who actually delivers. See the current rankings and put them head-to-head at **[Gangsta AI's best-AI breakdown](https://gangstaai.org/best-ai)**. Don't take my word for it — make them earn the yes.

Sources

  • [MarkTechPost — Meet the New Claude Opus 5 (July 24, 2026)](https://www.marktechpost.com/2026/07/24/meet-the-new-claude-opus-5-frontier-class-agentic-coding-and-computer-use-at-unchanged-opus-pricing/)
  • [Vellum — Claude
Leer artículo completo en dev.to