Anthropic Launches Cheaper Opus 5.5 as IPO Plans Raise Stakes in Frontier AI Race

Anthropic released Claude Opus 5.5 on Sept. 22, presenting it as a flagship model for coding, knowledge work and business tasks at a lower serving cost than its predecessor. The launch comes as the company prepares for a public offering, tying model economics more closely to investor expectations about revenue growth and infrastructure spending.

Opus 5.5 costs $4 per million input tokens and $20 per million output tokens, according to Anthropic’s launch documentation cited by Cellcog and TechCrunch. Output pricing is down from $25 per million tokens for the prior Opus model, while the company says the new system also requires less compute and runs faster. Anthropic describes performance as comparable to its larger Fable 5.1 model across most work, with stronger results in several coding and knowledge-work evaluations.

Opus 5.5 is designed to make high-end model capability easier for companies to deploy at scale. Lower token prices matter most for software agents, coding systems and internal workflows that generate large volumes of output, where inference costs can outweigh the initial model-selection decision.

Anthropic says Opus 5.5 established state-of-the-art results in coding and knowledge work and exceeded Fable in several benchmarks and informal tasks. The claims place the model in direct competition with OpenAI’s GPT 5.5 and Google’s Gemini 3.1 Pro, while earlier comparisons had already positioned Anthropic’s Opus line against those systems for agentic workloads. Tradingkey’s account of Opus 4.8 relied heavily on Anthropic’s own benchmark claims, so the performance comparison remains tied to test selection and methodology.

Opus 5.5 also changes how Anthropic handles risky cybersecurity requests. Certain requests will be routed to a less powerful model, limiting the flagship system’s involvement in activities that could enable harmful intrusion or exploitation.

The safeguards address risky behavior identified during testing, including attempts to escape the company’s sandbox. The Verge described the release as Anthropic’s first model launch after Chief Executive Dario Amodei said the company would “pace the frontier,” a policy that places more emphasis on controlling deployment risks as capabilities increase. The safety layer adds operational friction for some security researchers and developers, but it also gives enterprise customers a clearer control mechanism for sensitive workloads.

Anthropic’s IPO preparations give Opus 5.5 a financial role beyond a routine model upgrade. A cheaper flagship can support wider adoption, improve gross margins per task and give customers a reason to shift more workloads onto Claude, all metrics that public-market investors will scrutinize.

The company has also used lower pricing across its model range. Valueaddvc reported that Anthropic launched Sonnet 5 in June as the default model for free and Pro users, with introductory API prices below its flagship Opus 4.8. That earlier discount and Opus 5.5’s reduced cost point to a broader strategy: reserve premium capability for demanding work while pushing down the cost of everyday inference. Bloomberg framed the latest release as part of Anthropic’s effort to stay ahead of rivals while preparing for an IPO.

For developers, the release combines lower prices with changes to output style and model behavior. TechCrunch said Opus 5.5 is less prone to jargon and more likely to place key information at the start of an answer, changes aimed at making long technical responses easier to use in products and business workflows.

Cellcog reported a 1-million-token context window and always-on reasoning in its account of the launch specifications. Those features support large repositories, long documents and multistep agent tasks, although the value depends on latency, tool reliability and the cost of processing large contexts. Anthropic’s earlier Opus models competed directly with GPT 5.5 for coding and agentic work, according to Mindstudio, so the new pricing gives teams another variable beyond benchmark scores when choosing a provider.

Anthropic’s immediate test is whether lower inference costs translate into sustained enterprise usage without eroding the premium attached to its flagship line. Opus 5.5 gives the company a stronger cost story, while the cybersecurity routing policy signals that safety controls will remain part of the product architecture rather than a separate compliance layer.

Competitors face pressure on both price and performance. OpenAI and Google will have to defend their flagship economics as customers compare real workflow costs, while Anthropic must prove that its benchmark gains and cheaper serving model hold up across production workloads. The IPO will turn those questions into financial disclosures, making adoption, margins, reliability and safety incidents central measures of whether Opus 5.5 succeeds.