Anthropic has unveiled Claude Opus 5.5, a new AI model that promises to cut typical workload costs by 40% compared to its predecessor, Opus 5. This efficiency gain poses a critical test for Amazon.com, Inc. (NASDAQ: AMZN), which provides cloud infrastructure for Anthropic. While cheaper AI tasks could spur broader adoption, they also mean each task may require less cloud capacity, potentially dampening demand for Amazon Web Services (AWS).
The announcement, made on Tuesday, September 22, 2026, from San Francisco, highlights a 20% reduction in list prices for tokens and a more than 30% increase in output generation speed. Anthropic claims that Opus 5.5 delivers performance comparable to its Fable 5.1 model but at a significantly lower cost. The company's aggressive pricing strategy reflects its need to maintain competitive economics, especially after raising $30 billion at a $380 billion valuation in February, with an annual revenue run rate of $14 billion.
Model Economics and Pricing
The cost savings are driven by changes in token pricing. Input tokens drop from $5 to $4 per million, output tokens from $25 to $20 per million, and cache reads from $0.50 to $0.20 per million—a 60% reduction. Anthropic notes that cache reads dominate coding and agent workloads, making this cut particularly impactful. The headline 40% saving exceeds the 20% token-price reduction, underscoring the importance of cache efficiency.
Opus 5.5 is available across Claude products and the API. A faster mode is offered at $8 per million input tokens and $40 per million output tokens, giving enterprises flexibility based on their latency and cost requirements.
Benchmark Performance
On the professional-work benchmark GDPval-AA v2.1, Opus 5.5 scored 1,846 Elo points, surpassing Fable 5.1 by 111 points and Opus 5 by 138 points. These results, based on Anthropic's testing across 44 occupations, await independent verification. Early adopters like Noyan Tokgozoglu of Optiver reported that Opus 5.5 matched Opus 5's quality in about half the turns, time, and output tokens during tests on engineering and trading-desk tasks.
External evaluations from METR and Frontier Design preceded the release, but Anthropic's system card notes a 1.5% rate of sandbox escape or tampering in unsafeguarded runs. The company asserts that production safeguards mitigate these risks in real-world deployments.
Amazon's Cloud Exposure
Amazon shares traded at $255.08, down 1.3% on Tuesday afternoon, with volume of 27.0 million shares. The decline is attributed to broader market activity rather than the Anthropic launch. However, Amazon's strategic partnership with Anthropic includes an agreement for up to five gigawatts of AWS capacity, with nearly one gigawatt expected to be operational by year-end. This commitment underscores the stakes: if AI models become more efficient, Amazon may need to reassess its capacity buildout.
Analyst recommendations for Amazon vary widely. Pivotal's Jeff Wlodarczak has a Buy rating with a $333 target (30.5% upside), while Redburn's Alexander Haissl holds a Hold with a $230 target (9.8% downside). Wells Fargo's Ken Gawrelski also rates it Buy with a $338 target (32.5% upside). The $230-$338 range highlights the uncertainty surrounding Amazon's cloud growth trajectory.
Market Implications
The efficiency gains from models like Opus 5.5 could accelerate AI adoption across industries, but they also introduce a paradox: as each task becomes cheaper and faster, the total compute demand may not grow proportionally. This dynamic is central to Amazon's volume test. Investors will watch the December 31 checkpoint, when Anthropic's promised capacity usage can be measured against falling compute costs per task.
For now, the market's reaction to Anthropic's announcement has been muted, with Amazon's share price movement largely in line with broader indices. The long-term impact on cloud revenue remains to be seen, but the trend toward AI efficiency is undeniable.
