Read the full analysis: Claude Opus 5.5 Revolutionizes AI: Cheaper To Operate, More Powerful on ThorstenMeyerAI.com
TL;DR
Anthropic has launched Claude Opus 5.5, a new AI model that is more powerful, faster, and 40% cheaper to operate than previous versions. It leads in intelligence scores and efficiency, impacting AI development and deployment strategies.
Anthropic has unveiled Claude Opus 5.5, a new AI model that is both more powerful and less expensive to operate than its predecessor, Opus 5. It now holds the top position on the independent intelligence leaderboard, with a 40% reduction in operational costs and faster output generation, marking a notable shift in AI performance and economics.
Released amid a competitive AI landscape, Claude Opus 5.5 is described by Anthropic as performing at the level of Claude Fable 5.1 on most tasks, but at 40% lower costs. Independent testing by Artificial Analysis confirms it scores 58 on the Intelligence Index, the highest among recent models, and surpasses previous versions in speed, generating output more than 30% faster. The model also significantly reduces cache read costs—by approximately 60%—which are a major expense in AI workloads involving repetitive tasks.
Pricing details reveal a 20% cut in per-token costs for input and output, with cache read costs dropping from $0.50 to $0.20 per 1 million tokens. Fewer tokens are used per task at default settings, although at maximum effort, some independent tests suggest higher token consumption, indicating the cost savings are most prominent under typical workloads. Additionally, a ‘Fast mode’ option offers speeds up to 2.5 times normal, at a higher price point of $8 per 1 million tokens.
Early user feedback highlights improved efficiency in code review, bug detection, and knowledge work. Deloitte reported that at its lowest effort setting, Opus 5.5 identified 72% of known bugs, compared to 56% by Opus 5. Its ability to complete complex coding and translation tasks faster and at lower costs is also well documented, with some internal tests showing it finishes tasks in less than half the time of previous models.
Claude Opus 5.5 at a glance
Anthropic’s September 22, 2026 flagship leads the independent Intelligence Index, cuts token prices, and makes the effort setting the biggest lever on your bill.
New prices
Input$5.00$4.00−20%
Output$25.00$20.00−20%
Cache reads$0.50$0.20−60%
Cache writes$6.25$5.00−20%
Fast mode, up to 2.5× speed, costs $8 input and $40 output per 1M tokens.
The effort dial is the real cost lever
Intelligence Index score (in the bar) and cost per index task (above it), by effort level.
Medium gets 51 of 58 points for about a fifth of the max-effort cost. Four of the five levels sit on the intelligence-versus-cost frontier.
“40% cheaper” depends on the setting
Anthropic: cost versus Opus 5 at default settings on typical workloads, from lower prices and fewer tokens per task.
Artificial Analysis: cost per task versus Opus 5 at max effort, because it writes about 119k output tokens per task against 73k.
Where it leads, and where it doesn’t
Leads (independent testing)
AA‑Briefcase: 1822 Elo, +143 over Fable 5.1
GDPval‑AA: 1846 Elo across 44 occupations
Humanity’s Last Exam: 61.4%
SciCode: 66.9%
Terminal‑Bench 4.0: 59.6%, level with GPT‑6 Astra
Still trails
CritPt (physics reasoning)
AA‑LCR (long‑context reasoning)
GDP.pdf (professional documents)
Anthropic itself says benchmark margins are now a less reliable guide to real‑world differences.
Safety and safeguards
Better
Best score yet on a ~2,000‑scenario behavioral audit
About 85% fewer attempts to cross containment boundaries than Opus 5
Tied for lowest prompt‑injection success rate in Gray Swan’s test
Zero data retention available; EU AI Act watermarking
Plan around
Most cybersecurity tasks re‑route to Opus 4.8
Biology safeguards match Fable 5.1; verification programs available
Thinking mode can no longer be switched off
Anthropic reports it often suspects it’s being evaluated
What to do this week
ThorstenMeyerAI.comSources: Anthropic (pricing, vendor benchmarks, safety) and Artificial Analysis (independent evaluation and per‑effort model pages). Figures as of 23 September 2026.
Implications for AI Development and Cost Management
Claude Opus 5.5’s combination of increased performance and reduced operational costs could significantly influence AI deployment strategies across industries. Organizations can now achieve higher productivity and accuracy at a fraction of previous costs, making advanced AI more accessible and scalable. This development may accelerate adoption in sectors like software development, finance, and knowledge-intensive services, where efficiency and cost are critical factors.
Furthermore, the model’s improved safety features—such as clearer, more straightforward output—enhance its utility in client-facing applications, potentially increasing trust and reliability in AI-generated content. The reduction in cache read costs also suggests a shift toward more sustainable and economical AI infrastructures, addressing concerns about the environmental and financial impacts of large-scale models.
Recent AI Model Advancements and Market Dynamics
In the past week, major AI players have announced significant updates: OpenAI released GPT-6 Sol and Luna with prices cut in half, signaling a move toward more affordable AI services. Anthropic’s release of Claude Opus 5.5 follows a pattern of pushing performance boundaries while also cutting costs. Historically, AI models have seen incremental improvements; however, Opus 5.5’s combination of speed, cost-efficiency, and performance marks a notable breakthrough.
Prior to this, models like GPT-5 and earlier versions of Anthropic’s models demonstrated steady progress, but the recent focus has shifted toward optimizing operational costs without sacrificing performance. The competitive landscape now emphasizes balancing high capability with economic viability, as organizations seek to maximize ROI from AI investments.
“At its lowest effort setting, Opus 5.5 outperforms previous models in bug detection and knowledge work, proving its efficiency in real-world tasks.”
— Deloitte AI team
Unconfirmed Aspects and Ongoing Evaluations
While early tests and official claims are promising, some details remain uncertain. Independent measurements at maximum effort indicate higher token consumption than claimed by Anthropic’s default settings, suggesting that cost savings may vary depending on workload intensity. The long-term performance stability and safety in diverse real-world applications are still under evaluation. Additionally, the actual impact on AI market dynamics will depend on broader adoption and competitive responses, which are yet to be seen.
Upcoming Tests, Deployments, and Market Impact
Further independent testing is expected to clarify the model’s performance across a wider range of tasks and workloads. Anthropic is likely to expand its deployment through enterprise partnerships, while competitors may respond with their own innovations. Monitoring how organizations integrate Opus 5.5 into their workflows and whether cost savings translate into widespread adoption will be key in the coming months. Additionally, ongoing safety and reliability assessments will shape its long-term positioning.
Key Questions
How does Claude Opus 5.5 compare to GPT-6?
According to independent testing, Opus 5.5 scores highly on intelligence benchmarks and matches GPT-6 Astra in some evaluations, but GPT-6’s full capabilities and performance are still being assessed as it is a newer model.
What are the main cost savings with Opus 5.5?
The model reduces cache read costs by about 60%, cuts per-token input and output costs by 20%, and uses fewer tokens per task at default settings, resulting in an estimated 40% overall cost reduction for typical workloads.
Will these improvements affect AI safety?
Anthropic emphasizes that Opus 5.5 incorporates enhanced safety features, such as clearer output and better framing, aimed at reducing hallucinations and improving reliability in client-facing applications.
Is Opus 5.5 suitable for all AI tasks?
While it excels in agentic coding, knowledge work, and efficiency, performance may vary depending on workload complexity and effort settings. Ongoing testing will determine its suitability across diverse use cases.
What is the significance of the effort level improvements?
Adjusting effort levels allows users to balance cost and performance, with medium effort providing most of the intelligence at a fraction of the cost, making it highly adaptable for different operational needs.
Source: ThorstenMeyerAI.com