Anthropic Launches Claude Sonnet 5 and Claude Science to Drive Enterprise Agent Efficiency

Anthropic Launches Claude Sonnet 5 and Claude Science to Drive Enterprise Agent Efficiency

On June 30, 2026, Anthropic officially released Claude Sonnet 5, its most advanced workhorse agentic model to date. Positioned to bridge the cost-performance gap between standard models and the frontier Opus class, Sonnet 5 delivers near-flagship capabilities at a highly competitive price point. Parallel to the model release, Anthropic launched Claude Science, a specialized workbench designed specifically for researchers and pharmaceutical labs focusing on protein structure rendering and drug discovery.

Enhanced Agentic Reasoning and Cost Controls

Sonnet 5 is engineered to operate autonomously, plan multi-step workflows, and seamlessly interact with computer environments (such as browsers and terminals). According to Anthropic, the model narrows the capabilities gap with Opus 4.8 while undercutting competitor pricing:

• Pricing: Anthropic is offering introductory pricing of $2 per million input tokens and $10 per million output tokens through August 31, 2026, after which it will rise to a standard rate of $3 per million input and $15 per million output tokens. For comparison, Opus 4.8 is priced at $5 per million input and $25 per million output tokens.
• Effort Controls: Users can adjust the model's reasoning "effort" level. At medium effort, Sonnet 5 provides high cost efficiency¹, while at extra-high effort, its performance on complex benchmarks like BrowseComp (agentic search) and OSWorld-Verified (computer use) can match Opus 4.8.
• Autonomous Execution: Early access partners report that Sonnet 5 successfully completes complex multi-step coding, debugging, and data tasks that previously stalled mid-execution, and is highly capable of tracing failures to their root causes in messy "brownfield" codebases.

Automated Audits, Real-Time Cyber Safeguards, and Claude Science

Safety evaluations show that Sonnet 5 is generally safer to use in agentic contexts than its predecessor, Sonnet 4.6, with lower rates of hallucination and sycophancy. However, the model exhibits improved general intelligence that slightly raises its partial success rate on complex cybersecurity evaluations, such as developing exploits for Firefox browser vulnerabilities (evaluated in collaboration with Mozilla).

To mitigate potential dual-use risks, Anthropic has shipped Sonnet 5 with real-time cyber safeguards enabled by default. These safeguards detect and block malicious command execution in real time, though they are less restrictive than the strict safeguards deployed on Fable 5.

To further its commercialization push ahead of its anticipated IPO, Anthropic's Claude Science workbench aims to capture high-margin enterprise spending in the biotech and pharmaceutical sectors. The platform optimizes workflows for specialized scientific research, including genomic analysis and molecular modeling.