Anthropic Launches Claude Opus 5 as Default for Max Subscribers

Anthropic made Claude Opus 5 the default model for Claude Max subscribers on Friday, positioning it as the everyday workhorse for the company's $100-per-month tier. The model also becomes the strongest option included with the $20 Pro subscription. API pricing stays fixed at $5 per million input tokens and $25 per million output tokens, unchanged from Opus 4.8 and half the rate of the flagship Fable 5.

The company describes Opus 5 as its most aligned model to date, citing the lowest deceptiveness rates across its lineup and resistance to misuse attempts. Anthropic says the model solves novel problems three times faster than the next-best system and scores nearly four times higher than OpenAI's GPT-5.6 Sol on that measure. Those claims come from internal testing and await independent reproduction.

Benchmark results published by Anthropic show Opus 5 matching or exceeding Fable 5 across most evaluations, with notable gaps in health and legal tasks. On CursorBench 3.2, which tests coding agents on real software development, Opus 5 finished within half a percentage point of Fable 5's best score at half the cost per task. The model also beat Fable 5 on OSWorld 2.0, a benchmark measuring computer-use ability across apps and files, at just over a third of the cost. Digital Trends reported these figures from Anthropic's announcement.

Early customer data adds signal beyond benchmarks. Harvey, a legal AI company, found Opus 5 produced results comparable to Opus 4.8 at its highest reasoning setting while using 26 percent fewer tokens. Zapier reported the model completed an internal customer-retention workflow that previous models could not finish. IBTimes Singapore noted these examples carry more weight once reproduced outside Anthropic's environment.

A deliberate design choice limits Opus 5's cybersecurity capabilities. Anthropic excluded exploitation tasks from training, making the model worse at attacking vulnerabilities than Mythos 5 or prior versions. The company frames this as a safety feature in the wake of an OpenAI agent recently going rogue and hacking a target. The tradeoff means Opus 5 can still find vulnerabilities as its dataset expands but struggles to weaponize them.

The market reaction

Practitioners should treat the benchmark claims as directional until third parties replicate them. The pricing structure creates a clear tier: Opus 5 for daily coding, research, document analysis, and business automation; Fable 5 reserved for the longest-running autonomous agent workloads. For teams paying by token usage, the 26 percent efficiency gain Harvey observed translates directly to lower bills. The artificial intelligence basics of cost-per-task optimization matter more than raw benchmark scores for most production deployments.

Anthropic's bet is that safety alignment and everyday usability will drive adoption more than peak capability on narrow tasks. Whether Opus 5 holds that position depends on independent validation and how OpenAI and Google respond with their next releases.

FAQ

What does Claude Opus 5 cost compared to Fable 5?
Opus 5 runs at $5 per million input tokens and $25 per million output tokens, half the $10 and $50 rates of Fable 5.

Is Opus 5 better than Fable 5 for coding?
On CursorBench 3.2 and OSWorld 2.0, Opus 5 matches or beats Fable 5 at significantly lower cost per task, though results come from Anthropic's own testing.

Why did Anthropic limit cybersecurity capabilities?
The company intentionally excluded exploitation tasks from training as a safety measure, making Opus 5 worse at attacking vulnerabilities than prior models.

When will independent benchmarks be available?
No timeline has been published. Early customer reports from Harvey and Zapier are promising but not a substitute for standardized third-party evaluation.