Anthropic launches Claude Opus 5: near-flagship performance at half the price
The new AI model sits just behind the Fable 5 flagship – at the same price as its predecessor. The launch came three days before China's free giant model and intensifies the AI price war.
AI company Anthropic on Thursday launched Claude Opus 5, a new model marketed with the promise of "near Fable 5 intelligence at half the price". Fable 5 is the company’s most capable flagship model, and according to Anthropic’s own published benchmarks the new model sits close behind it – at half the cost per task.
The most striking part is the price tag: Opus 5 costs exactly the same as its predecessor Opus 4.8 – 5 dollars per million tokens in and 25 dollars out. Performance, however, has jumped sharply: on the company’s published benchmarks the model solves more than twice as many advanced coding tasks as its predecessor, and on OSWorld, which measures the ability to operate software and websites autonomously, it climbs from 55.7 to 70.6 per cent.
A price war from both ends
The launch came just three days before Chinese lab Moonshot AI on Sunday released the world’s largest open-weight AI model, Kimi K3, as a free download. Together, the two launches paint a picture of an industry in a full price war: open models push prices down from below, while the closed vendors respond by packing ever more performance under the same price tag.
For subscribers the change is immediate: Opus 5 is now the default model in Claude’s Max plan and selectable in the Pro plan, at no price increase. Developers got API access from launch day.
Fresher knowledge and a fast mode
The new model also has the freshest knowledge base in Anthropic’s lineup – it knows the world up to May 2026, against January for both the Fable 5 flagship and its predecessor. In practice that means fewer outdated answers about this year’s events, regulations and technology.
Anthropic is also introducing an optional "Fast mode" delivering roughly 2.5 times faster responses at twice the price – aimed at use cases where response time is critical, such as customer service and interactive AI agents.
The agent race
The gains in so-called computer use – the model’s ability to operate software on its own – are seen as key to the next phase of the AI race: agents that complete entire work tasks independently. American labs Anthropic and OpenAI have led that field so far, but with open challengers like Kimi K3 close behind in performance tests, the gap is narrowing there too.
