Claude Opus 5, released by Anthropic on July 24, 2026, delivers nearly identical coding performance to the more expensive Claude Fable 5 while costing 26% less per task, forcing engineering teams to reconsider their default AI model for everyday development work. The new model scores 61 on the Artificial Analysis Intelligence Index, narrowly the highest of any model, and runs at the same price as its predecessor, Opus 4.8, at $5 per million input tokens and $25 per million output tokens.

What Makes Opus 5 Different From Previous Claude Models?

Opus 5 introduces a feature that fundamentally changes how teams think about model selection: five effort settings that let you trade processing tokens for quality on a per-task basis rather than committing to a single model tier. These settings, labeled low, medium, high, xhigh, and max, allow the same model to run cheaply on routine edits and expensively on complex refactors. On the GDPval-AA v2 benchmark, Opus 5's effort levels span 407 Elo points, with output token usage ranging roughly 8 times from low to max effort.

The model also ships in a Fast mode that runs approximately 2.5 times faster than the default speed at twice the base price, offering $10 per million input tokens and $50 per million output tokens when latency matters more than cost. This flexibility means you are no longer choosing a model per task so much as choosing an effort level within a single model.

How Does Opus 5 Actually Perform on Coding Benchmarks?

On the benchmarks that matter most for coding agents, Opus 5 either matches Fable 5 or comes extremely close. On CursorBench 3.2 at maximum effort, it lands within 0.5% of Fable 5's peak score while costing half as much per task. On Frontier-Bench v0.1, an agentic software-engineering benchmark, Opus 5 surpasses every other model and more than doubles Opus 4.8's score at a lower cost per task. On Terminal-Bench v2.1, it hits 89% at maximum effort, roughly in line with the leader, GPT-5.6 Sol at xhigh effort.

The real-world cost advantage becomes clear when you measure cost per completed task rather than just token prices. Artificial Analysis reports that Opus 5 costs $2.03 per Intelligence Index task at maximum effort, compared to Fable 5's $2.75, the 26% saving that Anthropic and Artificial Analysis both lead with. However, Opus 5 is still more expensive than Opus 4.8 at $1.80 per task and Claude Sonnet 5 at $1.53 per task, so it is not automatically the cheapest answer for every workload.

When Should Teams Still Use Claude Fable 5?

Despite Opus 5's strong performance, Fable 5 remains the better choice for specific, narrower use cases. Teams should keep Fable 5 when three conditions are true: first, when factual precision is the product and hallucination rates matter. Artificial Analysis found that Opus 5 still trails Fable 5 on the AA-Omniscience knowledge benchmark, and although it gained 7 points over Opus 4.8 on accuracy, it answers more often when uncertain, pushing its hallucination rate up 14 points to 50%. For retrieval-heavy or compliance-sensitive answers, that regression matters significantly.

Second, teams should use Fable 5 when the horizon is very long and agents need to run autonomously for days. Third, Fable 5 remains ahead on frontier science tasks, where Artificial Analysis noted Opus 5 matches it on Humanity's Last Exam at 53% but sits behind the GPT-5.6 family on the CritPt physics evaluation. For a team spending on Fable 5 purely because it was the best coding model, however, Opus 5 is the cheaper default that gives up very little.

How to Choose the Right Claude Model for Your Team's Needs

• Everyday Engineering Work: Default to Opus 5 for in-IDE assistance, pull-request review, bug fixing, multi-file refactors, and terminal or DevOps automation. Opus 5 is within 0.5% of Fable 5 on CursorBench, joint first on the Coding Agent Index, and 26% cheaper per task, making it the model Anthropic now points its Max subscribers toward first.
• High-Volume or Budget-Constrained Agents: Use Claude Sonnet 5 as the value floor. At $1.53 per Index task, it is the cheapest option in the lineup, and for high-volume, latency-sensitive, or budget-capped agents it is often the right call even though it sits below the Opus tier on raw intelligence.
• Factual Precision and Compliance-Heavy Tasks: Keep Fable 5 for retrieval-heavy or compliance-sensitive answers where hallucination rates and factual accuracy are non-negotiable, since Opus 5's hallucination rate increased 14 points to 50% compared to Fable 5.
• Long-Running Autonomous Agents: Maintain Fable 5 for agents designed to run autonomously for extended periods, as Anthropic still positions its flagship for this use case.

The pricing structure also includes cache optimization opportunities. Cache writes carry a 25% premium at $6.25 per million tokens with a 5-minute time to live, and cache hits get a 90% discount at $0.50 per million tokens. An agent that keeps a large repository or system prompt warm can cut its real input bill well below the $5 headline price.

Anthropic's own framing is direct about the positioning. The company calls Opus 5 "a thoughtful and proactive model that comes close to the frontier intelligence of Claude Fable 5 at half the price." By holding the price flat from Opus 4.8 while lifting performance, Anthropic widened the set of workloads that are cheap enough to automate. The fact that Opus 5 is now the default model on Claude Max, which millions of subscribers use without changing a setting, signals where Anthropic expects the volume to shift.

For engineering teams currently paying for Fable 5 to drive everyday coding agents, the economics have shifted. Opus 5 gives you most of the same quality for a smaller bill, and the places to keep Fable 5 are narrower than they were before this release. The real win for teams comes at high and xhigh effort settings, where Artificial Analysis found Opus 5 can outperform both Opus 4.8 and Sonnet 5 at a lower cost per task, because it wastes fewer tokens getting to a correct answer. The migration lesson from past model swaps applies here too: the saving is real only if you also right-size the effort setting, or you will pay maximum-effort prices for medium-effort work.