**Anthropic** has released Claude Opus 5.5. The company says it reduces operating costs by about 40% compared with Opus 5 for token-based general workloads, and has also significantly improved response generation speed. At the same time, it has drawn market attention amid reports that the firm is aiming for a corporate valuation of up to around $2 trillion through a corporate initial public offering (IPO) in November.
Key points
Anthropic said that the general token-based charged workload cost of Opus 5.5 is about 40% lower than Opus 5.
At Terminal-Bench 4.0, it recorded 66.4%, and the output generation speed improved by more than 30%.
There are projections that Anthropic’s valuation in the November IPO could reach about $2 trillion, but the company has not yet finalized the terms.
Claude Opus 5.5 pricing structure
Anthropic said on September 22 that it launched Opus 5.5, the first model in its Claude 5.5 lineup. The main target use cases are code writing and refactoring, research, and professional knowledge-based work. The company emphasized that the model delivers performance close to Claude Fable 5.1 in most tasks while using fewer tokens than Opus 5.
API pricing is $4 per 1 million input tokens and $20 per 1 million output tokens, with both items reduced by 20% compared to Opus 5. Cache read costs also dropped by 60%, to $0.20 per 1 million tokens.
Anthropic estimates that if these fee adjustments and reductions in token usage are combined, the company’s typical execution costs could be reduced by about 40%. It also said output generation speed improved by more than 30%.
Performance metrics were also upgraded. The Terminal-Bench 4.0 score rose sharply to 66.4%, up from Opus 5’s 52.3%. In CursorBench 4.0, it reached 57.8%, and GDPval-AA v2.1 recorded 1,846 points. However, Anthropic warned that because the performance gap between models at the current level has narrowed, even small benchmark differences may not translate directly into real-world differences in day-to-day work.
Early customer case studies were also shared. One user completed a code migration of about 680,000 lines in less than a day, and another example had code auditing of 200,000 lines finished within three hours. According to Anthropic, the same auditing task with Opus 5 took more than 20 hours.
Opus 5.5 can be used alongside the Claude product lineup through Amazon Web Services, Google Cloud, and Microsoft Azure. Sonnet 5.5 and Haiku 5.5 from the same generation are expected to be released sequentially within the coming weeks.
Reference article: The US Economy Makes Billions From You Not Bothering, Says Citrini Research
Variables surrounding Anthropic’s IPO
This release came amid intensifying global debate over the pace of AI development at the frontier. Anthropic’s chief executive officer (CEO) **Dario Amodei** has previously proposed slowing the rate of capability improvements across the industry, and leaders including OpenAI CEO Sam Altman and Elon Musk have publicly supported the idea in part. By contrast, former U.S. President Donald Trump said on September 22 that the U.S. will not “stifle growth,” adding that it could respond by involving law enforcement agencies if necessary.
The **Wall Street Journal** reported that Anthropic is preparing for a November IPO, and that a scenario in which the company could raise up to $100 billion and be valued at roughly $2 trillion is under consideration.
However, these are still preliminary figures, and Anthropic has not released official offering terms. In particular, as the company pushes to expand enterprise demand and will need to gauge sentiment in the offering market, reducing inference costs and improving profitability could become key points to persuade investors.
The IPO timeline is also fluid. Earlier reports suggested the company would go public in October, but the latest report says the target date has been pushed to November in order to reflect the third-quarter results more fully. This suggests the timing of the listing, the amount to be raised, and the company’s valuation may be further adjusted depending on market conditions and performance going forward.
Further reading: Trump Renames AI “Super Intelligence” For U.S. Government Documents
