Anthropic has released Claude Opus 5, a new AI model that surpasses its previous flagship, Claude Fable 5, in performance across key benchmarks while offering a more cost-effective solution for businesses. Opus 5 is now available via API and subscription, stepping in to fill the operational role previously intended for Fable 5.
Performance Benchmarks Highlight Opus 5’s Superiority
Claude Opus 5 has demonstrated significant gains in performance. On the Frontier-Bench v0.1, which evaluates AI coding agents on end-to-end software engineering tasks, Opus 5 scored 43.3%. This marks a notable improvement over Claude Fable 5’s 33.7% and OpenAI’s GPT-5.6 Sol’s 34.4%. In tests assessing novel problem-solving on ARC-AGI-3, Opus 5 achieved 30.2%, substantially higher than GPT-5.6 Sol’s 7.8% (Fable 5 was not tested on this benchmark). The GDPval-AA v2 knowledge work benchmark further illustrates Opus 5’s strength, with an Elo rating of 1,861, compared to Fable 5’s 1,747 and GPT-5.6 Sol’s 1,736.
Fabian Hedin, as cited in materials shared by Anthropic, reported that Opus 5 shows a 22% improvement over Opus 4.7 in agentic coding tasks and offers greater stability with reduced variance. Internal evaluations by the developer platform Lovable and observations from Ultima Genomics suggest Opus 5 operates with enhanced capabilities, with Ultima Genomics noting its behavior as akin to a “careful scientist.”
Cost-Effectiveness and Business Value
For businesses, Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens. An accelerated “Fast” mode is available at double these rates. This pricing structure is presented as more economical for businesses than that of Claude Fable 5. Opus 5 is accessible to all users through a subscription service.
Addressing Fable 5’s Challenges
The introduction of Opus 5 follows a challenging period for its predecessor, Claude Fable 5. Launched on June 9, Fable 5 was withdrawn globally just three days later due to a U.S. government emergency export control order related to a jailbreak vulnerability. Although Fable 5 was reintroduced on June 30, it transitioned to a credits-only model. Opus 5 now assumes the primary role for paying users, offering a more stable and readily available option.
Anthropic’s AI model lineup includes tiers such as Haiku (fast and affordable), Sonnet (mid-range), Opus (a powerful workhorse), and Mythos (which includes Fable 5 and Mythos 5). Mythos 5 is a restricted version available through Project Glasswing for vetted cybersecurity researchers and critical infrastructure operators.
External Validation and Competitive Landscape
Beyond internal assessments, Opus 5 has also performed well in external evaluations. Zapier tested Opus 5 on its AutomationBench, where the model successfully completed a full business workflow, achieving 100% success. While Opus 5 demonstrates broad capabilities, Fable 5 reportedly retains a slight edge in specific legal and health-related tasks.
In the broader AI market, Moonshot AI, backed by Alibaba, recently unveiled Kimi K3, a 2.8-trillion-parameter open-weight model. Independent benchmarks position Kimi K3 third overall, behind Fable 5 and GPT-5.6 Sol, though it reportedly surpasses them in certain areas.
Anthropic’s Claude Opus 5 is now available, providing businesses with a more performant and cost-effective AI solution, while also addressing the operational continuity concerns that arose with the Fable 5 model.
Why This Matters
The materials describe a narrow update: Anthropic has launched Claude Opus 5, a new AI model that is cheaper for businesses to operate than Claude Fable 5 and achieves higher scores on key performance benchmarks. The exact date of Opus 5’s release.
Broader Context
Source materials place the factual news in this context: Anthropic’s lineup runs four tiers: Haiku (fast and cheap), Sonnet (mid-range), Opus (heavy workhorse), and Mythos (includes Fable 5 and Mythos 5).
Leave a Reply