Analysis
Anthropic released Claude Opus 5.5 this week at $4 per million input tokens and $20 per million output tokens -- 20% lower list pricing than Opus 5 -- with the company estimating typical workloads will cost about 40% less overall because the new model also uses fewer tokens to complete tasks, according to Anthropic and Inc42's coverage. Cache reads -- which Anthropic says make up the majority of costs on agentic and coding workloads -- are priced 60% lower, at $0.20 per million tokens.
The 'Pace The Frontier' Test Case
Anthropic is explicit that Opus 5.5 is its first model release since CEO Dario Amodei published "We Must Pace The Frontier" on September 12, an essay calling on AI companies to deliberately slow the pace of frontier capability development until safety and alignment research catches up. Rival CEOs -- OpenAI's Sam Altman, xAI's Elon Musk and Google DeepMind's Demis Hassabis -- publicly endorsed the essay within hours of its publication, and within a week a proposed federal antitrust class action had been filed in San Francisco accusing all four companies of using pacing as cover for an illegal Sherman Act agreement to slow AI development. Opus 5.5's launch is now a live test case for that argument: is this a company genuinely trading capability for safety margin, or a cost-and-efficiency release that would have shipped on roughly this timeline regardless of the essay.
“- OpenAI GPT-6 Sol -- $2/$10 per million input/output tokens, roughly half GPT-5.6 Sol's price.”
The Pricing Comparison
Three frontier labs cut API prices within days of each other this week:
- Anthropic Claude Opus 5.5 -- $4/$20 per million tokens list price, cache reads down 60% to $0.20/million, typical workloads about 40% cheaper overall.
- OpenAI GPT-6 Sol -- $2/$10 per million input/output tokens, roughly half GPT-5.6 Sol's price.
- xAI Grok 4.7 -- $2/$0.50/$6 per million tokens (input/cached input/output) below a 200K-token prompt.
Anthropic's steepest cut, to cache pricing specifically, is the most targeted of the three -- a bid to retain developers building coding and agentic workflows on Claude rather than a broad capability claim.
What Pacing Doesn't Resolve
A cheaper, faster, similarly capable model is genuinely ambiguous evidence for the pacing debate: it is consistent with a company that slowed its own roadmap for safety reasons, and equally consistent with a company optimizing for cost and speed because a larger capability jump simply wasn't ready this cycle. The antitrust lawsuit's core claim -- that the four companies' near-simultaneous public endorsements of pacing functioned as a coordinated agreement to slow competition -- remains unresolved and now sits in discovery, where internal roadmap documents could eventually settle the question this release cannot.
What To Watch
For startups building on Claude, the direct effect is a falling cost of goods sold, mirroring the same pattern now playing out across OpenAI and xAI's APIs. The more consequential signal for Anthropic's own strategy is whether its next release after Opus 5.5 is another efficiency-focused update or a genuine capability-frontier model -- that choice, more than anything in this week's pricing, will show what "pacing the frontier" actually means in practice.
