Anthropic has released Claude Opus 5, a new AI model that delivers near-flagship performance at half the price of its most expensive publicly available system. The model, released on July 24, costs $5 per million input tokens and $25 per million output tokens, compared to Fable 5's $10 and $50 pricing. Despite the lower cost, Anthropic's benchmarks show Opus 5 outperforming Fable 5 on eight of thirteen evaluation tests, signaling a major shift in how AI companies are competing on value rather than raw capability alone.

The release reflects a broader industry trend where the most expensive model is no longer necessarily the best choice for most users. As the AI market matures in 2026, companies and developers are increasingly matching their AI tool to the specific task at hand, rather than defaulting to the flagship system. Opus 5 is now the default model on Claude Max and the strongest model available on Claude Pro, accessible through the Claude API, Amazon Bedrock, Google Cloud, and Microsoft Foundry.

How Does Opus 5 Compare to Fable 5 on Real-World Tasks?

Anthropic's testing reveals that Opus 5 delivers exceptional performance on the kinds of work that matter most to developers and knowledge workers. On Frontier-Bench v0.1, an internal evaluation spanning physics, chemistry, and cryptography problems, Opus 5 scored 43.3 percent compared to Fable 5's 33.7 percent, while more than doubling Opus 4.8's score at a lower cost per task. On the ARC-AGI 3 problem-solving benchmark, Opus 5 scored roughly three times higher than the next-best competing model.

On CursorBench 3.2, a coding-focused evaluation, Opus 5 at maximum effort mode achieves a score within 0.5 percent of Fable 5's peak performance at half the cost per task. Cursor co-founder Sualeh Asif noted that "Claude Opus 5 delivers near Fable 5 intelligence at Opus speed and cost. On CursorBench it's just under Fable 5 and has many of the same behaviours". For scientific research tasks, Opus 5 shows particular strength, with a 10.2 percentage point improvement in molecular structure inference and a 7.7 point increase in protein-related tasks compared to its predecessor, Opus 4.8.

Sualeh Asif

"Claude Opus 5 delivers near Fable 5 intelligence at Opus speed and cost. On CursorBench it's just under Fable 5 and has many of the same behaviours," said Sualeh Asif, co-founder of Cursor.

Sualeh Asif, Co-founder at Cursor

What Practical Features Make Opus 5 Different?

Beyond raw benchmark scores, Opus 5 introduces several practical improvements that affect how developers and teams use the model. The system includes a full effort ladder with five settings: low, medium, high, xhigh, and max. This "effort" dial allows users to control how much computation the model dedicates to reasoning through a problem, checking its work, and exploring alternatives. Low effort is fast and cheap; max effort spends more tokens and time but produces better answers on difficult tasks.

Opus 5 also offers a Fast mode that runs at approximately 2.5 times the default speed for double the base price, useful when wall-clock time is the constraint. Thinking is enabled by default, though disabling it now returns an error at xhigh or maximum effort settings, so users who pinned specific model configurations should re-test before switching.

One significant advantage for security-adjacent development work is Opus 5's lighter cyber classifiers. The model permits vulnerability discovery in source code while blocking binary scanning, penetration testing, and exploit generation. Anthropic expects its classifiers to intervene roughly 85 percent less often on Opus 5 than on Fable 5, with flagged requests falling back to Opus 4.8. This reduction in friction is meaningful for teams doing legitimate security research without the overhead of constant safety interventions.

How to Choose Between Claude Models Based on Your Needs

• Everyday Coding and Knowledge Work: Opus 5 is the recommended choice for most developers and knowledge workers. It delivers near-flagship performance at half the cost, making it the strongest model available on Claude Pro and the default on Claude Max.
• Budget-Conscious Teams: Claude Sonnet 5 and Haiku 4.5 remain available for teams with tighter budgets. Sonnet 5 carries an introductory price of $2 per million input tokens and $10 per million output tokens through August 31, after which pricing may increase.
• Maximum Performance Tasks: Fable 5 remains available for teams that need absolute peak performance, though Anthropic's own data shows Opus 5 often delivers comparable results at half the cost. Fable 5 is included on Max plans and premium Team seats at up to 50 percent of weekly usage limits.
• Restricted Cybersecurity and Biology Work: Mythos 5, Anthropic's most capable and most tightly restricted system, remains available only to select partners for offensive cybersecurity and biological research tasks where Opus 5 deliberately trails.

What Does This Mean for the Broader AI Market?

Opus 5's release marks a turning point in how AI companies compete. In 2025, choosing a better AI model meant switching to a different system entirely. In 2026, it increasingly means adjusting an effort dial on the same model. Anthropic reports that Opus 5 at maximum effort lands within 0.5 percent of Fable 5's peak CursorBench score at half the cost per task, demonstrating that the dial is now a genuine cost lever.

The practical consequence is that "which model should I use?" has become "which model, at what effort?" This shift reflects a maturation of the AI market where cost per completed task matters more than raw capability rankings. Most people should start with free tiers, Anthropic suggests, since ChatGPT, Claude, Gemini, and Grok all offer credible free options. For those moving beyond free, the choice depends on matching the assistant to your ecosystem, the tier to the task, and the effort level to the difficulty.

Anthropic also confirmed that Opus 5 carries no data retention requirement for general access, consistent with prior Opus models. This contrasts with Fable 5, which retains inputs and outputs as a safety measure, a distinction that matters for teams with strict zero-retention policies. On alignment and safety, Anthropic's internal behavioral audit found Opus 5 to exhibit higher adherence to Claude's Constitution and lower incidents of deceptive behavior than previous models.

The release of Opus 5 comes as Anthropic and OpenAI both pursue stock market listings filed in June, with Wall Street scrutinizing their business performance more closely than ever. Cost efficiency has become a critical competitive factor, and Opus 5 directly addresses customer concerns about the expense of running advanced AI systems. By delivering near-flagship intelligence at half the price, Anthropic is signaling that the future of AI competition lies not in building bigger models, but in building smarter, more efficient ones that deliver better value per task.