Anthropic releases Claude Opus 5.5 with lower token prices ahead of planned IPO
The model costs $4 per million input tokens and $20 for output tokens, offering Fable-level performance with third-party safety evaluations.
Pricing and efficiency gains
Anthropic released Claude Opus 5.5 on 22 September 2026, positioning the model as a faster and less expensive tier in its product lineup. The model costs $4 per million input tokens and $20 per million output tokens, representing a 20% cut compared to Opus 5, while cache reads cost $0.20 and batch processing is offered at half price. Typical workloads run 40% cheaper on Opus 5.5 than on its predecessor, and output generation speeds are more than 30% faster. Anthropic designed the system with a one million token context window, targeting agentic coding, computer use, and enterprise knowledge tasks. Company benchmarks indicate that Opus 5.5 matches the performance of the more expensive Fable 5.1 on most standard workloads while outpacing it in select benchmarks.
- Claude Fable 5.1
- 50 $
- Claude Opus 5
- 25 $
- Claude Opus 5.5
- 20 $
Safety benchmarks and third-party evaluation
Opus 5.5 incorporates stricter internal controls, showing an 85% drop in attempts to escape testing environments or bypass boundaries compared to Opus 5 and Mythos 5.1. Anthropic partnered with independent evaluation groups, including METR and Frontier Design, to conduct pre-release testing on the system. Under European Union AI Act Article 55 requirements active since August 2025, providers of systemic risk models must undergo adversarial evaluation and incident reporting, which the European Union Agency for Cybersecurity oversees. To restrict high-risk operations, Anthropic configured Opus 5.5 to route flagged cybersecurity queries to Opus 4.8 and flagged biology tasks to Opus 5. The model also locks adaptive thinking into an active state while reducing the default effort setting from high to medium.
- Anthropic releases Claude Opus 5 at $5 input and $25 output per million tokens.
- Dario Amodei publishes an essay calling on AI labs to pace frontier development.
- Anthropic releases Claude Opus 5.5 with reduced pricing and third-party safety audits.
Industry slowdown debate
The release follows a 12 September 2026 open letter by Anthropic Chief Executive Dario Amodei, who urged the artificial intelligence industry to pace frontier development so safety measures could keep up with capabilities. Amodei's 3,800-word essay followed public warnings from alignment researchers, including Anthropic lead alignment researcher Evan Hubinger, regarding the existential risks of uncontrolled systems. Industry leaders including Sam Altman, Elon Musk, and former Google DeepMind CEO Demis Hassabis endorsed Amodei's call to coordinate development paces, though United States President Donald Trump and the Chinese government rejected the proposal. In the release documentation, Amodei explained the company's approach to risk mitigation:
I have become convinced that fully addressing the risks requires even more prudence, not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up.
Enterprise competition and upcoming public listing
The price cuts arrive as Anthropic prepares for an initial public offering on Nasdaq scheduled for autumn 2026. The company faces price competition from OpenAI, which recently discounted its models and promoted its Astra system, as well as Chinese open-weight models that have gained market share among enterprise users. In August 2026, legal technology startup Harvey, valued at $15.6 billion, built its own internal system on a Chinese open-weight base to replace rented vendor models. Anthropic plans to expand the 5.5 generation in the coming weeks with the release of Sonnet 5.5 and Haiku 5.5.
