Anthropic chose this week to do two things at once: cut the price of its most capable model and push back its public debut. Both moves, coming within days of each other, say the same thing about the company’s immediate priority: revenue now.
On Tuesday, Anthropic released Claude Opus 5.5, the first model in its Claude 5.5 family. The company said the model matches the performance of its Claude Fable 5.1 system on most tasks while costing 40 percent less to run than Claude Opus 5 and producing output more than 30 percent faster.
The pricing follows the same direction. Anthropic set input at $4 per million tokens and output at $20 per million tokens, 20 percent below Opus 5, with cache hits priced at 20 cents per million tokens, a 60 percent reduction. The cuts arrive as the company courts the enterprise customers whose workloads will determine how fast its revenue compounds.
The model is available across all major cloud platforms, including Amazon Web Services, Google Cloud and Microsoft Azure, a distribution push that matters for a company trying to turn a research lead into durable market share. Anthropic said Claude Sonnet 5.5 and Claude Haiku 5.5 will follow in the coming weeks, carrying many of the same efficiency and safety improvements.
The release is the latest turn in a pricing war that has defined the AI industry for two years. Anthropic, OpenAI and Google have each been cutting the cost of inference while raising the capabilities of their frontier models, betting that cheaper tokens will expand the market faster than they shrink the margin on each one.
Anthropic was founded in 2021 by Dario Amodei and his sister Daniela Amodei, who left OpenAI along with a group of researchers concerned about how fast the field was moving. The company has since raised enormous sums from Amazon and Google, whose clouds also happen to host the model it released this week, a relationship that binds Anthropic’s funding to its distribution.
Amazon has committed billions of dollars to Anthropic and carries Claude through its Bedrock service, while Google has invested alongside and offers the model on its own platform. Those ties mean each price cut Anthropic announces flows immediately into the catalogs of the two largest cloud providers, multiplying the discount’s reach.
Anthropic has named its models after literary figures, a choice that sets it apart from rivals whose names read like version numbers. Fable, the system Opus 5.5 is measured against, sits at the top of that hierarchy, and the company has used the comparison to signal that its mid-tier model now does much of what its flagship did only months ago.
The same week carried news of the financial stakes. The Wall Street Journal reported that Anthropic’s initial public offering has slipped a month, from October to November, and that the company is targeting a valuation of about $2 trillion. The company’s revenue run rate reached $47 billion as of May, and it has told investors to expect roughly $190 billion to $200 billion in revenue by 2028, according to people familiar with its finances.
Cutting prices and delaying a listing in the same week is not a contradiction. A company that lowers the cost of its flagship model is bidding for usage, and usage is what underwrites the growth story an IPO needs. The delay gives Anthropic another month of operating numbers to show bankers and a market that has grown wary of paying for AI’s future before it arrives.
Anthropic has positioned the 5.5 family around efficiency rather than raw capability, a nod to the customers who complain that frontier models are fast enough but too expensive to run at scale. The company said the same safety work that went into Opus 5 carries into the new line, though it offered few specifics on how the cheaper model behaves under stress.
Enterprise customers have been the company’s fastest-growing segment, and their complaints have shaped the 5.5 line. Many say they run Claude constantly and watch the bill rather than the benchmark. A model that answers as well but costs 40 percent less to run, and nearly free on repeat queries, speaks directly to that buyer.
The discount on cache hits, which let repeat queries run against stored context rather than fresh computation, is aimed squarely at the agentic workloads that dominate enterprise spending. Agents ask the same system the same question many times over; making those repeat calls nearly free changes the economics of building on top of Claude.
Whether Opus 5.5 can hold its own against Fable 5.1, its own higher-end system, is the question the market will answer through usage. Anthropic’s claim is that most work does not need the frontier model, and that a cheaper model with near-frontier performance will pull volume from rivals without cannibalizing the top of its lineup.
The stakes for the IPO are plain. Anthropic must convince investors that a company spending enormous sums on computing can still show a path to profit, and price cuts make that argument harder in the short term. The company is wagering that volume will make up the difference. November will tell whether the market agrees.
