Anthropic launched Opus 5 on July 24, 2026, an artificial intelligence model with performance close to Fable 5, the firm's most advanced model, at half the cost. The company charges $5 per million input tokens and $25 per million output tokens.
According to Ars Technica, Anthropic's move responds to growing market pressure: companies like Cursor and Meta are building routers that automatically select between models based on the task, avoiding the use of costly models like Fable 5 for every query. Open-source models such as Kimi K3, from Chinese firm Moonshot, also offer comparable performance at $15 per million output tokens. For Mexican companies and developers already running Anthropic models in production, the difference is material: every query, every line of generated code, and every automated analysis costs roughly half what Fable 5 charges.
Dianne Penn, Anthropic's Chief Product Officer, told Forbes México that the company continues to deliver frontier intelligence consistently and aims to make it as accessible as possible. In internal benchmarks such as Frontier-Bench and DeepSWE, Opus 5 matches or slightly outperforms Fable 5 on programming tasks, and surpasses both its predecessor Opus 4.8 and OpenAI's GPT-5.6-Sol model across nearly all categories. Unlike its more advanced models, Anthropic did not train Opus 5 using advanced offensive cybersecurity techniques, which reduces the risk of misuse without sacrificing vulnerability-detection capability. Opus 4.8 had been launched in May 2026; Fable 5 arrived in June of the same year.
Opus 5 is now available through the Anthropic API and the Claude application. For Anthropic, token efficiency (greater capability at the same or lower price) defines the competitive battlefield for generative AI in the coming months, as open models and automatic routers erode the margin for charging a premium on raw capability.
This article was written with artificial intelligence assistance from verified sources and reviewed by a human editor before publication.