Anthropic's Claude Sonnet 5 offers agentic AI near Opus 4.8 performance at lower cost, as Meta, Amazon and Uber cut back on AI token spending.
Anthropic has released Claude Sonnet 5, a new AI model built to handle autonomous, multi step tasks at a lower cost than the company's top tier systems, arriving at a moment when businesses are actively trying to rein in how much they spend on AI usage.
A Cheaper Path to Agentic AI
Sonnet 5 sits below Anthropic's most capable models, Claude Fable 5 and Mythos 5, in raw power. But the company says it closes much of that gap while staying well ahead of its own predecessor, Sonnet 4.6, and landing close to the performance of Opus 4.8. In a statement, Anthropic said testers found that Sonnet 5 completes complicated jobs that earlier Sonnet versions would abandon partway through, and that it reviews its own work without being told to. The company frames that combination, thorough task completion plus self checking, as arriving at what it calls an attractive price point for agentic work, the kind of AI activity where a system takes a series of actions on its own rather than just answering a single prompt.
Why Companies Are Suddenly Watching Token Spending
The timing lines up with a broader pullback across the tech industry. Heavy AI usage, measured in tokens (the units that represent chunks of text a model processes), had become expensive enough that even large firms started reining it in. Meta, Amazon, and Uber have each taken steps to curb unnecessary token consumption, a practice some in the industry have started calling tokenmaxxing. That shift toward cost discipline is exactly the kind of environment where a model like Sonnet 5, cheaper but still capable of finishing complex tasks, becomes more appealing to businesses watching their AI bills.
Security Add Ons and a Recent Ban
Anthropic also built cybersecurity safeguards into Sonnet 5, even though the model itself was not specifically trained for security tasks. The company says these protections mirror what it already uses in its Opus 4.7 and 4.8 models. That detail carries extra weight given what happened earlier this month, when the Trump administration required Anthropic to block foreign nationals, both inside and outside the United States, from using its most powerful models, Fable 5 and Mythos 5, over concerns they could be turned into hacking tools. Anthropic had to pull those models offline for that group of users, a move that reportedly frustrated customers relying on them.
An IPO Waiting in the Wings
Sonnet 5 lands as Anthropic prepares for a stock market debut expected later this year. The company confidentially filed paperwork with the Securities and Exchange Commission on June 1. OpenAI followed with its own confidential filing a week later, and is separately working with the Trump administration to expand access to its newer GPT 5.6 Sol, Terra, and Luna models. Both companies are effectively racing toward public markets while also negotiating how far the US government will let their most advanced systems travel beyond American borders.
