Anthropic just made its best everyday AI far cheaper to run. On Friday the company launched Claude Opus 5, a model it says nearly matches its most capable public system while costing half as much, 5 dollars per million input tokens against 10. For the businesses and developers who pay by the token, that gap is the difference between an AI feature they can afford to ship and one they cannot.
The system Opus 5 is measured against is Fable 5, Anthropic’s top publicly available model, built for long autonomous jobs an agent can run with little supervision. Opus 5 does not try to beat it. It aims to get close for the ordinary work most people actually do, the knowledge tasks and automation they reach for all day, at a price that makes reaching for it an easy call. Anthropic says it comes near Fable 5’s ability for half the money.
The company leans on benchmarks to make that case. On the coding and knowledge tests it cites, Opus 5 set new highs, and on one measure of controlling a computer it beat Fable 5’s best result at about a third of the cost. The exception is security, where Anthropic says its restricted Mythos 5 model still ranks first. Opus 5 is already the default on the top Claude Max plan, so most paying users are meeting it now whether they went looking or not.
There is a quieter win buried in the fine print. Unlike Fable 5, Opus 5 does not hold on to what you type and what it replies for 30 days. Anthropic keeps that month of data for its frontier models as a safety measure, but for a company routing sensitive work through the model, the plainer point is what lands. Your inputs are not sitting on Anthropic’s servers for a month. Set next to the lower price, that makes Opus 5 the easier model to say yes to.
Opus 5 arrives in a crowded few weeks. OpenAI shipped its GPT-5.6 family earlier this month, and Google is readying Gemini 3.5 Pro, though Bloomberg reported that launch has slipped and Google disputes it. The pressure is not only American. Last week China’s Moonshot AI released its openly available Kimi K3, which the lab says nearly matches both Fable 5 and OpenAI’s GPT-5.6 Sol, a reminder that the lead the top US labs hold keeps shrinking.
Washington is watching all of it. The Trump administration first made Anthropic pull Mythos 5 and Fable 5 over security fears, then let the company restore them, while OpenAI handed its GPT-5.6 Sol to regulators for review before releasing it widely. The labs are trying to ship fast and stay on the government’s good side at the same time, and that tension now shapes every big launch.
Underneath the model race is a race to Wall Street. Anthropic filed confidentially to go public in June and looks on track to reach the market later this year, with OpenAI close behind but possibly waiting until 2027 as it fights an Apple lawsuit accusing it of stealing trade secrets. Seen that way, Opus 5 is as much a business move as a technical one. A near top tier model most customers can afford to run every day is exactly what widens a paying base right before an IPO.
Disclosure: Entrelligence uses Anthropic’s Claude tooling in its editorial workflow.
Your First 10 AI Skills
10 practical AI skills, copy-paste prompts and a 7-day plan to start using AI with confidence.
Download the guide →