Anthropic has launched Claude Opus 5, its latest frontier model that delivers performance close to the company’s most powerful systems on many key tests while costing roughly half as much to run. The model went live on July 24 and is already available through Anthropic’s chatbot, API, and major cloud platforms.
The company positions Opus 5 as the practical everyday choice for developers, enterprises, and knowledge workers who need strong reasoning and coding ability without paying top-tier rates. It becomes the default model for Claude Max subscribers and the strongest option included with Claude Pro plans.
How Opus 5 Stacks Up on Benchmarks
Anthropic reports that Claude Opus 5 comes close to the capabilities of Claude Fable 5, its higher-priced frontier model, across a wide range of domains. On several internal and public evaluations, it matches or exceeds Fable 5 while using fewer resources.
On Frontier-Bench, a test of end-to-end software engineering tasks, Opus 5 more than doubled the score of its predecessor Opus 4.8 and outperformed other leading models at a lower cost per completed task. On CursorBench 3.2, which measures real coding agent performance, it landed within half a percentage point of Fable 5’s peak result at maximum effort, yet at roughly half the cost per task.
The model also posted strong results on knowledge-work and problem-solving tests. On ARC-AGI-3, which focuses on novel puzzles rather than memorized patterns, Opus 5 scored three times higher than the next-best model. On Zapier’s AutomationBench, it achieved about 1.5 times the pass rate of competing systems at the same cost level. Computer-use benchmarks such as OSWorld 2.0 showed similar gains, with Opus 5 clearing Fable 5’s best score at roughly one-third the budget in some settings.
Independent evaluations have placed the model near the top of composite intelligence rankings and agentic knowledge-work leaderboards, often ahead of or level with more expensive rivals on cost-adjusted metrics.
Pricing That Changes the Calculation
API pricing for Claude Opus 5 stays at $5 per million input tokens and $25 per million output tokens, the same rates as Opus 4.8. That puts it at about half the per-token cost of Fable 5, which runs higher. Anthropic also notes it is slightly cheaper than some comparable OpenAI frontier offerings on output tokens.
An optional Fast mode, currently in research preview, roughly doubles generation speed for twice the standard price. Users can also adjust an effort dial that trades some peak intelligence for lower token use and reduced cost on simpler tasks. At lower effort settings, the model still preserves much of its capability while consuming fewer resources.
This pricing structure makes high-end performance more accessible for daily production work rather than reserving it for occasional complex jobs.
Where the Model Excels in Practice
Developers highlight meaningful gains in agentic coding. Opus 5 handles multi-file features, large refactors, and end-to-end tasks more reliably than earlier Opus versions. It verifies its own work more carefully, iterates when needed, and produces fewer incomplete stubs. Multi-agent coordination shows fewer conflicts, and code-review performance finds real bugs with lower false-positive rates even at reduced effort levels.
Knowledge workers benefit from stronger performance on long-running professional tasks. The model maintains accuracy across research, analysis, and business automation workflows while using tokens more efficiently. Visual understanding and computer-use capabilities also improved, supporting more reliable interaction with interfaces and documents.
Safety features received attention as well. Anthropic describes Opus 5 as its most aligned Opus model yet and the least susceptible to being tricked into misuse. Cybersecurity guardrails are calibrated more tightly than on higher-risk models, with fewer unnecessary activations for legitimate security research or code scanning.
Availability and Practical Rollout
Claude Opus 5 is live across Anthropic’s platforms, including the Claude chatbot and API. It is also available on Amazon Bedrock with zero data retention by default in multiple regions, and through other enterprise channels. The 1-million-token context window carries over from recent Claude models, supporting longer documents and multi-step projects.
For Claude Max users, the switch to Opus 5 as the default happens automatically. Pro subscribers gain access to it as their strongest included model. Teams already running production workloads can adopt it without changing existing pricing tiers.
What This Means for Users and the Market
The release continues a pattern of rapid iteration. Anthropic has shipped multiple Claude 5-series models in a short window, focusing less on single blockbuster launches and more on steady gains in capability, efficiency, and cost. Opus 5 sits in the sweet spot for organizations that need near-frontier intelligence without the premium of the absolute top tier.
Riders who previously reserved Fable 5 for only the hardest jobs can now run more of their daily work on Opus 5 and still see strong results. Teams watching token spend closely gain a clearer path to higher performance without proportional budget increases. Independent testing will continue to refine the picture, but early benchmark data and developer feedback point to a model that delivers meaningful value at its price point.
Anthropic’s decision to keep pricing flat while lifting performance underscores a shift toward making advanced models practical for everyday use rather than experimental showcases. For many developers and enterprises, Claude Opus 5 now offers the closest combination of capability and cost efficiency currently available in the Claude lineup.