Claude Opus 5 launches with stronger coding skills, huge context, and the same API price as Opus 4.8
Anthropic has officially launched Claude Opus 5, its newest flagship AI model, positioning it as a major step forward for coding, reasoning, and advanced knowledge work. The model became available on July 24 across the Claude API, Amazon Bedrock, Google Cloud, Microsoft Foundry, and the Claude apps.
Anthropic is presenting Opus 5 as more than a routine upgrade. The company claims the new model outperforms competing AI systems in key areas such as software development and complex professional tasks, while offering better value than its higher-priced Fable 5 model.
Who can use Claude Opus 5?
Access depends on which Claude plan you use. Claude Max subscribers now get Opus 5 as the default model. Claude Pro users can select it as the most capable option available in their plan. Free users, however, remain limited to Sonnet 5 and do not receive access to Opus 5.
That makes Anthropic’s strategy clear: its most powerful AI model is reserved for paying customers. While some rival models are easier to test without a subscription, Anthropic is keeping its flagship capabilities behind premium access.
Claude Opus 5 pricing: what “half the price” really means
Anthropic’s claim that Opus 5 costs half as much needs some context. The model is not cheaper than its direct predecessor, Opus 4.8. Through the API, Claude Opus 5 costs 5 dollars per million input tokens and 25 dollars per million output tokens, which is the same pricing as Opus 4.8.
The “half-price” comparison applies to Fable 5, Anthropic’s more expensive model tier. In other words, Opus 5 delivers flagship-level performance at a lower cost than Fable 5, but it is not a discounted replacement for Opus 4.8.
There is also a Fast mode for users who need quicker responses. However, Fast mode costs twice the standard rate and is currently available only through the Claude API.
For most everyday users, token pricing will not matter directly because Claude subscriptions are billed as monthly plans rather than per-token charges. Developers and businesses using the API, however, will need to pay close attention to token usage, especially with larger prompts and long outputs.
Claude Opus 5 now uses reasoning automatically
One of the biggest practical changes in Claude Opus 5 is how it handles reasoning. With Opus 4.8, deeper internal reasoning had to be enabled manually. With Opus 5, reasoning is active by default.
The model can now decide when it needs to think more deeply and how much effort to spend on a task. This is designed to improve accuracy, planning, coding performance, and complex problem-solving without requiring users to adjust settings every time.
Anthropic gives users five reasoning levels: low, medium, high, xhigh, and max. High is the default setting. However, the company recommends using low or medium more often for routine tasks because those levels can still produce useful results while consuming far fewer tokens.
This matters because stronger reasoning can increase usage. Claude Opus 5 tends to provide longer answers by default and may check its own work without being asked. That can improve reliability, but it also uses more of a subscriber’s quota. Pro users who often run into usage limits may want to reduce the reasoning level for simpler prompts.
A one-million-token context window comes standard
Claude Opus 5 includes a one-million-token context window. That is both the default and the maximum, with no smaller version available. The model can also generate outputs of up to 128,000 tokens.
In real-world use, this gives Claude Opus 5 the ability to process extremely large documents and projects in a single session. Users can feed it long contracts, full technical manuals, large research files, extensive codebases, or multi-document project folders without splitting everything into smaller parts first.
For developers, legal professionals, analysts, writers, researchers, and enterprise teams, this could be one of the most important upgrades. A larger context window means the model can maintain awareness of more information at once, reducing the need for repeated summaries or fragmented workflows.
Safety improvements and fewer cybersecurity refusals
Anthropic says Claude Opus 5 is its best-aligned model so far. According to the company’s internal testing, it has the lowest measured rate of deceptive behavior among its models.
At the same time, Anthropic has loosened some restrictions around cybersecurity-related requests. The company says Opus 5 should trigger security refusals around 85 percent less often than Fable 5. This change appears aimed at making the model more useful for legitimate security research, defensive analysis, and professional cybersecurity workflows.
If Claude apps block a request, Anthropic says Opus 4.8 may automatically take over in certain cases. The company also notes that when it comes to generating attack code, Opus 5 still trails behind Mythos 5 in its own testing.
Independent testing is still needed
While Anthropic’s claims around Claude Opus 5 are ambitious, the strongest performance results currently come from the company’s own benchmarks. Independent comparisons against leading models such as OpenAI’s GPT-5.6 and major Chinese AI systems are still needed before the model’s real position in the market becomes clear.
For now, Claude Opus 4.8 remains available across supported platforms, giving users the option to continue with the previous model while testing the new flagship.
Claude Opus 5 arrives as a powerful upgrade for users who need advanced coding help, long-context analysis, stronger reasoning, and enterprise-ready AI performance. Its automatic reasoning, massive context window, and premium positioning make it one of Anthropic’s most important releases yet, though its full impact will depend on real-world use and independent evaluation.
