Anthropic has released a new flagship artificial intelligence model, Claude Opus 4.5, which the company says sets a new performance standard in coding and knowledge-intensive tasks. The announcement was made on Tuesday, with Anthropic claiming the model surpasses even larger competing systems across a range of benchmarks, including informal real-world tasks that rival models were unable to complete.

Opus represents the top tier in Anthropic's three-level Claude model family. Sonnet occupies the middle position, offering a balance between capability and cost, while Haiku serves as the fastest and most affordable option. The Opus line is designed for users who need the highest level of reasoning and task completion, and version 4.5 continues that positioning while introducing meaningful cost reductions.

Lower Prices and Faster Performance

One of the more notable aspects of this release is its revised pricing structure. Output tokens with Opus 4.5 will be billed at $20 per million tokens, a reduction from the $25 per million charged for the previous generation. Input and other usage metrics have seen comparable price decreases. The company attributes these lower costs to improvements in computational efficiency, meaning the model requires less processing power to run at scale.

Speed has also improved alongside the cost reductions. Anthropic says the model is faster to serve than its predecessor, a benefit that flows directly from the same efficiency gains that enabled the pricing changes. For developers and enterprise customers running large volumes of queries, the combination of lower cost and faster response times represents a meaningful operational improvement.

Changes to Communication Style

Beyond raw performance, Anthropic has made deliberate adjustments to how Opus 4.5 communicates with users. The model has been trained to rely less on technical jargon and to lead with the most important information at the beginning of its responses. These changes reflect ongoing efforts by AI developers to make their systems more accessible and practical for a broad range of users, not just those with deep technical backgrounds.

Rapid Release Cadence

The launch of Opus 4.5 comes approximately two months after Anthropic released Claude Opus 4 on July 24. The company's release pace has accelerated considerably over the past year, with incremental model improvements arriving at shorter intervals. In the same announcement, Anthropic confirmed that Sonnet 4.5 and Haiku 4.5 - the next tiers in the lineup - are expected to follow in the coming weeks, each bringing similar performance and efficiency improvements relative to their current versions.

Safety Classifications and Capability Thresholds

Anthropic has placed Opus 4.5 under the same safety framework as its more advanced Fable model, citing comparable capabilities in biology and cybersecurity. The company says the new model performs at a level similar to Fable in those domains, which triggers a corresponding set of safeguards. Those restrictions govern how the model can be used in sensitive areas, including limiting its ability to assist in identifying software vulnerabilities in compiled programs or providing meaningful support toward the development of biological weapons.

The application of these safeguards reflects Anthropic's tiered approach to model deployment, where more capable systems are subject to stricter usage controls regardless of which product line they belong to. The company has been developing this safety classification framework over several model generations as a way to match deployment restrictions to actual capability levels rather than applying blanket policies across all models.

A Shift in Development Philosophy

Opus 4.5 is the first model Anthropic has released since CEO Dario Amodei publicly endorsed the idea of deliberately moderating the pace of AI capability development. Earlier this month, Amodei published a statement outlining his evolving position on how the industry should approach the relationship between advancing capabilities and ensuring safety measures can keep up.

"I have become convinced that fully addressing the risks requires even more prudence - not just investing in risk prevention, but pacing the rate of capabilities advancement so that risk prevention has time to keep up."

This statement marks a notable shift in tone from a company that has long been at the forefront of pushing AI capabilities forward. Amodei's remarks suggest Anthropic is now weighing the speed of its development against the readiness of its safety infrastructure in a more formal and deliberate way.

Safety Training and External Evaluation

The safety training process for Opus 4.5 followed a broadly similar structure to previous releases. That process includes alignment testing and pre-release evaluation by external organizations, including METR and Frontier Design, which conduct independent assessments of model behavior before public deployment. These third-party evaluations are designed to catch potential risks that internal testing might miss and to provide an additional layer of accountability.

However, Anthropic noted that more sophisticated training and evaluation systems are already being developed for future model generations. The company said these systems will include enhanced security protocols and more advanced monitoring capabilities, though specific details have not yet been disclosed.

The company also addressed the broader role of public policy in AI safety, indicating that it views government and regulatory frameworks as an increasingly important part of the safety ecosystem as models become more capable.

"As AI becomes more capable, public policy should play a larger role in making sure the systems people rely on are safe. That capacity takes time to build, and we've started to put the infrastructure in place to support it."

Anthropic added that further details on its policy and infrastructure efforts would be shared in the near future. The statement signals that the company is preparing for a regulatory environment that may become more active as AI systems grow more powerful and their applications become more consequential across sectors ranging from healthcare and finance to national security.

With Opus 4.5 now available and two additional model tiers expected to follow in the coming weeks, Anthropic is positioning itself for a period of continued growth while publicly committing to a more measured approach to how quickly it advances the frontier of AI capability.