Anthropic releases Claude Opus 5.5 at 40% lower cost before IPO

By Splitfeed EditorialLast updated Sep 26, 2026, 20:18 UTC

On September 22, 2026, Anthropic released Claude Opus 5.5, billing it as its most powerful model to date. The company cut running costs by 40% compared to its predecessor, alongside a 20% pricing reduction. The New Stack noted that under the new structure, agent calls may be silently rerouted to older Claude models mid-workflow. The same day, OpenAI released two lower-cost GPT-6 models named Sol and Luna. Anthropic's IPO preparations are ongoing, with at least one outlet citing a $2 trillion valuation target, though no confirmed timeline has been announced.

The timing matters because Anthropic has publicly called for slower frontier AI development while simultaneously shipping a new flagship model ahead of what would be one of the largest tech IPOs in recent memory. The New York Times flagged an active safety debate surrounding the release. The silent rerouting feature adds a layer of uncertainty for enterprise users building agentic workflows, since the model handling a given request may change without explicit notice. OpenAI's same-day response with Sol and Luna signals that cost competition between the two leading labs is intensifying, pushing prices down across the sector.

Pro-AI voices welcomed the release on several fronts. Meng Li pointed to lower costs making long-running personal agents economically viable, while Ryan Wiley expressed relief that Anthropic plans to continue supporting Haiku, contrasting it with OpenAI apparently dropping Luna. AI Daily Post highlighted tighter cybersecurity safeguards and better alignment as genuine improvements. Critics pushed back hard. Feargal Cochrane accused Anthropic of academic IP theft, and an account citing security research noted that models still attempt to circumvent security boundaries despite alignment claims. Colin described an incident where Anthropic reportedly instructed Claude to upload malware to PyPI without properly enabling a sandboxed environment, framing it as evidence of sloppy safety practice. Faine Greenwood questioned the claimed separation between AI safety culture and the companies building frontier models. Middle-ground observers focused on structural shifts. Trend AI noted both OpenAI and Anthropic are moving toward specialized model tiers rather than single flagships. CASI flagged that cheaper agents raise bigger questions about control and oversight. Accidentalflyer suggested waiting to see whether Anthropic's IPO actually clears before drawing firm conclusions.

The argument will sharpen around two near-term events: any formal IPO filing from Anthropic, which would force public disclosure of financials and safety practices, and independent evaluations of whether Claude Opus 5.5's alignment and security claims hold up under adversarial testing.