Anthropic launches Claude Sonnet 5 — near-Opus performance at lower cost

The artificial intelligence landscape is often characterised by the release of massive, highly capable models that demand significant computational resources and come with a premium price tag. However, the practical application of these tools in everyday business environments often requires a different balance: robust capability paired with cost efficiency.

Anthropic's release of Claude Sonnet 5 on 30 June 2026 addresses this need directly, offering performance levels that approach their top-tier Opus models but at a fraction of the cost.

This development is particularly significant for teams in architecture, engineering and construction. Many firms have hesitated to deploy advanced agentic workflows for document analysis, cost estimation and programme management due to the prohibitive costs associated with the most powerful models. Sonnet 5 alters this calculus, making sophisticated automation accessible to a much broader range of projects and organisations.

Historically, developers and enterprise users have had to choose between the speed and affordability of mid-tier models and the advanced reasoning capabilities of frontier models. The previous generation, Sonnet 4.6, was widely adopted for its efficiency but often struggled with highly complex, multi-step tasks that required sustained focus.

Sonnet 5 represents a substantial upgrade. According to Anthropic's benchmark testing, the new model performs closely to Opus 4.8 across various evaluations, including reasoning, coding and knowledge work. It is specifically designed to be highly agentic, meaning it can autonomously make plans, utilise tools such as web browsers and terminals, and execute extended workflows without requiring constant human intervention.

This level of autonomy was previously the exclusive domain of larger, more expensive models. By bringing these capabilities to the Sonnet tier, Anthropic has effectively democratised access to advanced artificial intelligence agents. Early access partners have already reported substantial improvements in their workflows.

The true value of Sonnet 5 lies in its cost-performance ratio. Key pricing details:

• Introductory pricing (through 31 August 2026): $2 per million input tokens, $10 per million output tokens

• Standard pricing (from 1 September 2026): $3 per million input tokens, $15 per million output tokens

• Opus 4.8 comparison: $5 per million input tokens, $25 per million output tokens

• Cost savings: Organisations can achieve 40–60% cost reductions compared to Opus 4.8 for many workflows

For enterprises running high-volume tasks, such as processing extensive project documentation or managing automated customer service workflows, this price reduction can translate into significant operational efficiencies. Users can now adjust their "effort levels" to find the optimal balance between cost and performance for specific tasks. For medium-effort tasks, Sonnet 5 provides exceptional cost efficiency, while its high-effort performance can match Opus 4.8 in certain scenarios.

As models become more autonomous, ensuring they operate safely and reliably becomes increasingly critical. Anthropic's pre-deployment safety evaluations indicate that Sonnet 5 exhibits a lower rate of undesirable behaviours compared to its predecessor. It is more adept at refusing malicious requests and resisting prompt injection attacks, making it safer for deployment in agentic contexts where the model interacts directly with external systems.

Interestingly, the evaluations also revealed that Sonnet 5 has a significantly lower ability to perform advanced cybersecurity tasks, such as developing software exploits, when compared to the Opus models. While it can handle routine defensive cyber tasks, its limitations in offensive capabilities reduce the overall risk profile of the model.

Despite this lower risk profile, Anthropic has launched Sonnet 5 with real-time cyber safeguards enabled by default, consistent with the protections applied to the Opus tier. This proactive approach ensures that the model remains secure even as it executes complex, multi-step operations.

The enhanced capabilities of Sonnet 5 open up new possibilities for project delivery professionals. The model's ability to maintain context over long interactions and autonomously check its own work makes it well-suited for tasks that require meticulous attention to detail.

For example, when reviewing complex engineering drawings or cross-referencing multiple technical specifications, Sonnet 5 can trace issues to their root causes rather than simply addressing surface-level symptoms. Its proficiency in coding and debugging also makes it a valuable asset for teams developing custom software tools or integrating various project management platforms.

"Claude Sonnet 5 gives our agents a strong execution layer for multi-step software engineering work," noted Zimu Li, a member of the technical staff at a leading technology firm. "It handles sustained coding, tool use, and debugging well across messy technical contexts, and has been especially useful for workflows where follow-through and technical grounding matter."

The model's improvements in agentic performance have been validated across multiple use cases.

This shift in capability allows teams to redeploy their most experienced professionals to higher-value work, such as architectural decisions and strategic planning, rather than routine debugging and code review. The efficiency gains compound across organisations deploying the model at scale.

• Sonnet 5 delivers performance comparable to the more expensive Opus 4.8 model, particularly in agentic tasks involving reasoning and tool use.

• The competitive pricing structure significantly lowers the barrier to entry for deploying complex automated workflows at scale.

• Improved safety mechanisms make the model more reliable for autonomous operations, with lower rates of hallucination and susceptibility to prompt injection.

• The model's ability to execute multi-step plans and self-correct makes it highly effective for detailed technical analysis and coding tasks.

• Firms can now achieve high-quality outputs with fewer operational steps, improving overall efficiency in project delivery.

To understand how the shifting economics of artificial intelligence models can impact your project delivery strategies, subscribe to the Project Flux newsletter.

All content reflects our personal views and is not intended as professional advice or to represent any organisation.