Anthropic launched Claude Opus 5.5 on Tuesday, introducing a new model line that reduces costs by 40% on typical workloads compared to its predecessor, Claude Opus 5.

The release targets performance parity with previous top-tier models while requiring less compute. Anthropic stated the model generates responses faster and includes significant changes to subscription structures, removing five-hour usage caps for Pro, Max, Team, and seat-based Enterprise subscribers.

Pricing and access changes

The new pricing structure positions Opus 5.5 as a more cost-efficient option for high-volume users. Cache reads are priced significantly lower than standard input tokens, offering potential savings for repeated queries.

A faster mode is available at higher token rates. The removal of five-hour caps addresses a key constraint for heavy users, while a new rate-limit reset feature has been added to the subscription plans.

Safety and capability benchmarks

Anthropic highlighted safety improvements alongside performance gains. External groups, including Frontier Design and METR, evaluated Opus 5.5 before launch. An automated behavioral audit produced the strongest results the company has recorded to date.

The model is described as less likely to take difficult-to-reverse actions than newer models and more resistant to prompt injection than Opus 5. Capabilities in sensitive areas such as biology and cybersecurity include safeguards similar to those used for Claude Fable 5. Vetted organizations can apply to the Life Sciences Verification Program, with expanded access to the Cyber Verification Program planned for coming weeks.

Performance in agentic tasks

Benchmark results show Opus 5.5 performing at or near the top in categories including agentic coding, computer use, and knowledge work. Early testers reported completing large-scale code migrations and audits in hours rather than days. The model also identifies software performance bottlenecks while making fewer changes to application behavior than Opus 5.

One early tester noted the model's writing capabilities align closely with their own style. Anthropic cautioned that small differences in benchmark scores do not necessarily translate into meaningful differences in real-world performance.

Future releases

Anthropic plans to release Claude Sonnet 5.5 and Claude Haiku 5.5 in the coming weeks. These updates will extend the new model family beyond the flagship Opus offering, signaling a broader refresh of the company's product lineup.

Big Short investor Steve Eisman claims AI executives are manufacturing a "doomsday crisis" to protect their businesses, as competitive pressure mounts on Anthropic to release a new model.

Eisman argued on Friday that companies like Anthropic and OpenAI cannot afford to slow down despite public calls for caution, citing hundreds of billions of dollars in commitments to hyperscalers. He suggested the push for regulation is designed to establish barriers to entry and preserve pricing power against cheaper alternatives.

Competitive Pressure on Anthropic

Anthropic is evaluating a new model aimed at countering OpenAI’s GPT-6 Astra, according to Reuters. This development comes days after CEO Dario Amodei called for the industry to slow improvements in AI capabilities, a stance backed by OpenAI CEO Sam Altman.

The competitive landscape is shifting rapidly. OpenAI’s Astra accounted for about 13% of enterprise AI spending tracked by Ramp, compared with 8% for Anthropic’s Claude Fable. Additionally, OpenAI overtook Anthropic last week on OpenRouter, marking the first time it led in spending there in more than two-and-a-half years.

Meta Platforms Inc (NASDAQ: META), one of Anthropic’s largest customers, is looking to reduce its reliance on Anthropic models as it builds internal AI capabilities. Meta recently launched Muse, its first personal AI agent, powered by its most capable model to date.

Eisman’s Moat Argument

Eisman stated there are "no pricing moats" in the business, noting that open-weight models allow developers to download and run systems independently. He expects these models to drive a price war, challenging the sustainability of proprietary systems.

He warned that if Anthropic or OpenAI were to fail, the broader AI infrastructure buildout, including Nvidia Corp (NASDAQ: NVDA), could face significant disruption. Nvidia is considering investing up to $10 billion as an anchor investor in Anthropic’s planned IPO.

IPO Details and Market Sentiment

Anthropic is preparing what could be the largest IPO ever, potentially raising $100 billion at a valuation around $2 trillion. This would roughly double its $965 billion valuation from a May funding round.

Polymarket traders give Anthropic a 78% chance of going public by December 31, with about $3.7 million traded on the timing market. Reuters reported Anthropic may wait until after November’s U.S. midterm elections to list.

What the Numbers Show

Anthropic’s annualized revenue run rate exceeded $65 billion by the end of July, up from roughly $9 billion at the end of 2025. This rapid growth contrasts with Eisman’s warning of impending price wars driven by open-weight models and competitors like OpenAI. The divergence between high market confidence in a successful listing and the intensifying competitive pressure highlights the risk to future margins if pricing power erodes.