Anthropic's Claude Opus 5.5 lands cheaper, safer, and 3.6x busier

Anthropic dropped Claude Opus 5.5 on September 22, 2026, the first member of its new Claude 5.5 family. The company says it matches Fable 5.1, its pricier tier, while undercutting Opus 5 on cost. Translation: more muscle, smaller bill.

The math is blunt. Input tokens run $4 per million, output tokens $20 per million, roughly 20% below Opus 5. Cache reads took the bigger haircut, sliding 60% and landing at $0.20 per million tokens. Stack the list price cuts with that cache discount and Anthropic reckons typical workloads come out about 40% cheaper overall. The model ships through the Claude API, Amazon Web Services, Google Cloud and Microsoft.

Customers apparently noticed. Token volume climbed 3.6 times after launch, a hint that cheaper per-task costs are luring in fresh activity rather than just shaving invoices. Anthropic's bet is that volume more than covers the lower unit price — and that every extra token flowing through its cloud partners means more workloads on their platforms too, spread across all three providers rather than one.

Safety got the headline treatment as well. Anthropic commissioned outside evaluations from groups including Frontier Design and METR, which pointed to better alignment and containment than earlier models. The flashiest number: Opus 5.5 is reportedly about 85% less likely to try dodging predefined safety protocols than prior versions. That matters as enterprises hand agents access to codebases and internal systems, where a model that stays inside its lane is easier to sell to a risk committee.

The launch arrives alongside Anthropic's S-1 filing, the paperwork that precedes going public. The company reported $16.3 billion in revenue for the first half of 2026, up roughly 1,300% year over year. In the second quarter it posted an operating profit of $974 million and an adjusted operating margin of 19%. Latent Space framed the swing as a turnaround from a -126% loss to a 19% GAAP profit in a year, calling Anthropic's revenue and research efficiency among the highest in history.

In a Latent Space conversation, Anthropic engineering head Katelyn Lesse and product head Angela Jiang described how the company ships models — largely by using Claude to build Claude products — and a deliberate design call that keeps the systems controlling agents separate from the systems executing code. Next up: Sonnet 5.5 and Haiku 5.5 are expected to follow. With the S-1 public, investors will be watching whether that usage surge shows up in the next revenue and margin numbers.