Claude Opus 5 Arrives: Anthropic's Fourth Model Launch in Two Months

News Summary

Anthropic officially introduced Claude Opus 5 on Friday, July 24, 2026 (Pacific Time), positioning the new model as its best-performing and most cost-effective release to date. The company describes Opus 5 as nearly matching the capability of its top-tier Claude Fable 5 model while costing half as much per token, and it is now set as the default model for most users across Anthropic's consumer and developer products. This marks the fourth model Anthropic has shipped in under two months, following Claude Mythos 5, Claude Fable 5, and Claude Sonnet 5, all released in June 2026.

Pricing and Availability

Claude Opus 5 is priced at $5 per million input tokens and $25 per million output tokens, identical to its predecessor, Claude Opus 4.8. That means developers and enterprise customers get a substantial capability upgrade without any increase in cost. By comparison, Anthropic's most capable model, Claude Fable 5, carries roughly double the per-token price, making Opus 5 the more economical choice for teams that need strong reasoning without the premium price tag. The model is available immediately through the Claude API, Claude Platform on AWS, and supported cloud partners including Amazon Bedrock and Google Cloud Vertex AI.

Benchmark Performance

Anthropic highlighted several independent and internal benchmark results to demonstrate the leap in capability from Opus 4.8 to Opus 5:

• On FrontierBench v0.1, an agentic terminal-coding benchmark that evaluates real software engineering work such as multi-file changes, debugging, and building features from a specification, Opus 5 scored 43.3 percent at maximum reasoning effort. That compares with 18.7 percent for Opus 4.8, 33.7 percent for Fable 5, and 37.5 percent for a leading competing model.
• On ARC-AGI-3, a benchmark designed to test novel problem-solving in interactive, turn-based environments with no stated rules or goals, Opus 5 reached 30.2 percent — nearly four times higher than the closest tracked competitor model. The benchmark is specifically constructed to resist memorization and simple pattern-matching, rewarding models that can reason through unfamiliar situations step by step.

Anthropic said the gains stem largely from improvements in how the model plans and verifies its own work, rather than from raw scale alone.

New "Effort" Control for Balancing Cost and Capability

One of the most notable additions in Opus 5 is a user-facing "effort" setting, which lets developers choose between low, medium, and high levels of reasoning depth for a given task or prompt. Lower effort settings complete simpler requests faster and more cheaply, while higher settings allocate more computation toward planning, self-checking, and iteration on harder problems. Anthropic said this gives developers fine-grained control over the trade-off between response speed, cost, and accuracy, rather than forcing a single fixed behavior for every request.

Judgment and Self-Verification

Anthropic emphasized that Opus 5 is "much stronger at verifying its work and iterating carefully until it succeeds," a shift the company attributes to training the model to catch its own logical errors during planning rather than only after producing an answer. In practice, this means Opus 5 spends more of its reasoning process checking whether an intermediate step is correct and why an answer works, rather than simply producing output and moving on. Anthropic frames this as a step toward more reliable agentic behavior — useful for tasks like multi-step coding projects, data analysis, and workplace automation where a single unnoticed mistake can compound across many steps.

Context in Anthropic's Recent Model Roadmap

Opus 5 arrives as part of a rapid cadence of releases from Anthropic over the summer of 2026. In June, the company introduced Claude Sonnet 5, aimed at balancing speed and intelligence for everyday development work, alongside Claude Fable 5, its most capable and expensive model for the hardest reasoning and long-horizon agentic tasks, and Claude Mythos 5, a specialized variant made available through a limited research initiative. With Opus 5 now filling the mid-to-upper tier of the lineup at unchanged pricing, Anthropic has effectively refreshed its entire model family within a two-month window, a pace the company says reflects faster internal training and evaluation cycles.

Why It Matters for Developers and Educators

For students, researchers, and developers exploring applied artificial intelligence, Opus 5's combination of stronger reasoning benchmarks and unchanged pricing lowers the barrier to experimenting with more capable agentic systems — including coding assistants, automated research tools, and multi-step task agents — without a jump in operating cost. The new effort control also offers a concrete, hands-on example of how modern language models can be tuned to trade computation for accuracy, a concept increasingly relevant across machine learning coursework and applied AI engineering.