Core Summary
Anthropic has recently released a new model, Opus 5, which performs on par with its most advanced model, Fable 5, yet at half the price. This move marks a shift towards a “high cost-performance” strategy by this closed-source AI giant. Opus 5 not only continues the company’s tradition of offering “superior performance at an affordable price” but also includes features that allow users to customize costs and control the level of intelligence, while maintaining a strong focus on security. The behind-the-scenes rationale is to compete with Chinese open-source models and the open-weight camp. On the eve of its IPO, Anthropic aims to retain corporate customers by adopting a layered approach: using a quasi-flagship model to appeal to the general market and its high-end models to safeguard its technological leadership and security.
1. Opus5: Getting Nearly Flagship Intelligence for Half the Price
The most immediate advantage of Opus 5 is its incredible cost-performance ratio. Its pricing remains the same as its predecessor, Opus 4.8 (5 dollars per million text blocks input, 25 dollars for output), yet its performance is comparable to that of the more expensive Fable 5 (10 dollars for input and 50 dollars for output).
For example, in programming tests, Opus 5 and Fable 5 have only a 0.5% difference in score, but Opus 5 costs half as much to complete the same tasks. In software engineering tests, it outperforms all other models, doubling the performance while keeping costs lower. The same applies in life sciences: for instance, it is 10.2 percentage points more accurate in predicting molecular structures and 7.7 percentage points more accurate in predicting protein mutation impacts. In short, you get nearly flagship-level performance at a mid-range price.
2. A Gradual Shift towards Cost-Performance
The price reduction of Opus 5 is not an accidental decision but part of Anthropic’s strategy over the past year:
- October 2025: The lightweight model Haiku4.5 was launched at an extremely low price (1 dollar for input, 5 dollars for output), offering coding capabilities on par with the mid-range model Sonnet4, marking the first attempt to offer “superior performance at a lower cost.”
- November of the same year: The flagship model Opus 4.5 saw its price reduced by two-thirds (from $15/75 to $5/25).
- June 2026: The mid-range model Sonnet5, priced at $3/15, outperformed the then-flagship model Opus 4.8 in knowledge work tests.
With Opus 5, Anthropic has taken this strategy to the next level by providing quasi-flagship performance at a mid-range price.
3. Giving Users Control: Choose Between Intelligence and Cost Savings
Opus 5 includes a new “Effort” setting that allows users to customize the level of intelligence they desire. Want ultimate intelligence? That’s possible, but it may come at a higher cost and longer waiting times. Looking to save money? You can reduce the level of fidelity and still complete more tasks than with other models. Even the lowest-tier version of Opus 5 outperforms other models on the market in terms of the number of tasks it can handle.
Anthropic positions Opus 5 as a model for “daily, frequent use” and has made it the default option for its premium service, Claude Max. It’s like buying coffee: you can choose between high-quality, freshly brewed coffee (high intelligence) or instant coffee (low cost), with the choice ultimately in your hands.
4. Uncompromising on Security
Despite focusing on cost-performance, Anthropic has not neglected security. Some key aspects of Opus 5’s security design include:
- Reduced number of security triggers: It intercepts 85% less sensitive content than Fable 5, but it still does not allow everything.
- Prohibition of high-risk operations: While it can help identify vulnerabilities in source code, it cannot be used to perform binary file scans or penetration tests.
- No support for network attacks: Since it has not been trained for attack tasks, even if it detects vulnerabilities, it cannot be utilized as an attack tool.
- Continued emphasis on biosecurity: Similar to the previous version, Opus 5 will not be misused in high-risk biological research.
In other words, Opus 5 is “user-friendly without being dangerous”; for high-risk applications (such as network attacks or sensitive biological experiments), the more stringent flagship model Mythos5 is still required.
5. The Motivation Behind the Price Reduction
Anthropic’s decision to reduce prices is largely due to external pressures:
- The rise of Chinese open-source models: Many companies find closed-source models too expensive and are turning to free or low-cost open-source alternatives.
- The competition from the open-weight camp: Rivals have made their code publicly available, allowing companies to deploy them at lower costs.
On the eve of its IPO, Anthropic needs to retain customers by addressing their various needs. Opus 5 (quasi-flagship at a mid-range price) covers the daily requirements of most businesses, while Fable 5 (flagship) and Mythos 5 (limited versions) safeguard its technological leadership and security.
This approach creates a three-tiered strategy: affordable options for the general market, professional solutions for high-end customers, allowing Anthropic to both generate revenue and maintain its core competitiveness.
In Conclusion
Opus 5 represents Anthropic’s balanced approach: using cost-effectiveness to capture the market and leveraging security and advanced models to protect its position. In an increasingly competitive AI landscape, this closed-source giant is adapting to stay ahead with more flexible strategies.