Claude Opus 5 Outscores Fable 5 on Most Benchmarks—At Half the Value

Claude Opus 5 is out today. It’s cheaper for companies to run than Anthropic’s main mannequin, Claude Fable 5, which the corporate had positioned because the on a regular basis frontier product for paying customers. What’s extra, Opus 5 additionally outperforms it on vital benchmarks.

To know the place Opus 5 suits: Anthropic’s lineup runs 4 tiers. Haiku is quick and low cost. Sonnet is mid-range. Opus is the heavy workhorse. Above that sits the Mythos class—a tier Anthropic launched this spring—which incorporates Claude Fable 5 for the general public, and Claude Mythos 5, a model with fewer restrictions reserved by Challenge Glasswing for vetted cybersecurity researchers and important infrastructure operators.

Fable 5 has had a tough run because the subscriber flagship. It launched June 9, was pulled globally three days later after the U.S. authorities issued an emergency export management order citing a jailbreak vulnerability, and came back June 30—solely to shift instantly to a credits-only mannequin, now not included in customary plans. Opus 5 now fills the slot Fable 5 could not maintain.

Lovable, a developer platform with thousands and thousands of customers, ran Opus 5 on its inner evaluations and famous the features lengthen past uncooked scores: “It is not simply higher on our hardest agentic coding duties, up 22% over Opus 4.7, it is steadier, with far much less variance run to run,” Fabian Hedin mentioned in a press release shared by Anthropic.

The benchmarks

It might sound unusual, however Opus beats Fable on virtually every little thing that can matter to the on a regular basis consumer whereas not being labeled as Mythos-class like Fable.

On Frontier-Bench v0.1—a benchmark that checks whether or not AI coding brokers can full actual software program engineering duties end-to-end, scored as a share of duties handed—Opus 5 hit 43.3%. Fable 5 got here in at 33.7%. OpenAI’s GPT-5.6 Sol, Anthropic’s important business rival, scored 34.4%.

The widest margin is on ARC-AGI-3, a check of real problem-solving constructed round novel puzzles a mannequin could not have memorized from coaching knowledge, scored as a share of puzzles solved. Opus 5 hit 30.2%; GPT-5.6 Sol scored 7.8%; and Fable 5 wasn’t examined in any respect. On GDPval-AA v2—a information work benchmark scored by way of Elo scores, the chess-style rating system used to measure relative efficiency on actual skilled duties—Opus 5 reached 1,861 in opposition to Fable 5’s 1,747 and GPT-5.6 Sol’s 1,736.

Zapier examined Opus 5 on AutomationBench, an analysis that scores whether or not a mannequin can carry a full enterprise workflow from begin to end with out human assist. Their verdict: the mannequin “took a uncooked account-health workbook and ran a full churn-prevention sequence finish to finish: flagging at-risk accounts, alerting the appropriate proprietor, and summarizing for retention ops. Earlier fashions did not cross; Opus 5 hit 100%.”

Anthropic can be pitching Opus 5 as a analysis improve. Ultima Genomics, a DNA sequencing firm, mentioned the mannequin “behaves extra like a cautious scientist than any mannequin we have run. It reaches for the appropriate statistical checks to rule out confounders, cross-checks its personal outcomes by impartial strategies, and stays on monitor by lengthy multi-step analyses.”

That mentioned these two areas—authorized and well being—are the one ones by which Fable 5 excels by a tiny margin.

The discharge lands every week after Moonshot AI, a Beijing-based startup backed by Alibaba, unveiled Kimi K3—a 2.8-trillion-parameter open-weight mannequin (that means anybody can obtain the underlying code to run it independently) that Moonshot describes because the world’s largest open AI system. Impartial benchmarks persistently place Kimi K3 third total, behind each Fable 5 and GPT-5.6 Sol, beating these two in particular areas.

Opus 5 is offered now by way of API at $5 per million enter tokens and $25 per million output. (Tokens are the fundamental unit of data an AI mannequin can course of in each enter and output). A Quick mode operating at roughly 2.5 occasions the default pace can be out there, at twice the bottom value—$10 per million enter tokens and $50 per million output.

This launch might finish the anxiousness over Fable 5’s lack of public availability. Opus can be out there by way of subscription for everybody.