Anthropic Unveils Claude Sonnet 4.5 as Top Coding AI Model

Anthropic has officially launched Claude Sonnet 4.5, positioning it as "the best coding model in the world" for building production-ready applications rather than just prototypes, according to the company's official announcement.

The model demonstrates significant advancements in autonomous coding capabilities, with Anthropic reporting that it can maintain focus for "more than 30 hours" on complex, multi-step tasks during enterprise trials. This represents a substantial leap from previous models, which typically required human intervention after shorter durations. Anthropic AI researcher David Hershey confirmed this capability to TechCrunch, noting the model's ability to "build an application, stand up database services, purchase domain names, and perform a SOC 2 audit" without human oversight.

Claude Sonnet 4.5 achieves state-of-the-art results on the SWE-Bench Verified coding benchmark, which evaluates real-world software engineering abilities. The model maintains a 77.2% score on this benchmark, according to Anthropic's documentation, significantly outperforming previous versions and competing models. Anthropic emphasizes that the model's improvements extend beyond raw benchmark scores to practical application, with enterprise customers reporting "state-of-the-art coding performance" on longer horizon tasks.

Pricing remains consistent with the previous Sonnet 4 model at $3 per million input tokens and $15 per million output tokens. This pricing strategy positions Sonnet 4.5 as the optimal balance between capability and cost for developers, with Anthropic noting it offers "the sweet spot between capability and cost." The company reports that major enterprises like Apple and Meta have adopted Anthropic's models internally, with Anthropic's API access powering developer tools including Cursor, Windsurf, and Replit.

Anthropic's documentation highlights that Sonnet 4.5 represents a significant advancement in "computer use" capabilities, leading the OSWorld benchmark for real-world computer tasks at 61.4%—a 19.2% improvement over Sonnet 4's 42.2% score from four months prior. The model also demonstrates improved domain-specific knowledge and reasoning across finance, law, medicine, and STEM fields compared to older models like Opus 4.1.

Enterprise adoption has been particularly strong in developer tooling, with Cursor CEO Michael Truell stating that Sonnet 4.5 "represents state-of-the-art coding performance, specifically on longer horizon tasks." Windsurf CEO Jeff Wang echoed this sentiment, calling it a "new generation of coding models." Anthropic reports that Sonnet 4.5 has reduced vulnerability intake time for security agents by 44% while improving accuracy by 25%, directly benefiting enterprise security workflows.

Unlike previous models, Sonnet 4.5 incorporates "hybrid reasoning" capabilities that allow it to toggle between fast execution for routine tasks and extended thinking for complex logical challenges. This architectural improvement enables the model to generate internal reasoning content blocks before producing final responses, significantly improving output quality for difficult problems.

The launch follows Anthropic's broader 2025 model family release, which included the more expensive Opus 4.5 (achieving 80.9% on SWE-Bench Verified) and the cost-effective Haiku 4.5. While Opus 4.5 remains the flagship model for maximum capability, Sonnet 4.5 is positioned as the optimal choice for developers seeking the best balance of performance and cost for coding tasks.

Anthropic's approach contrasts with recent competitive developments, as OpenAI's GPT-5 has reportedly challenged Anthropic's dominance in coding benchmarks. However, Sonnet 4.5's demonstrated ability to handle "complex, codebase-spanning tasks" better than previous models positions it as a strong competitor in the enterprise AI coding space, particularly for organizations prioritizing autonomous agent capabilities over raw speed.

Artūras Malašauskas is an AI Systems Integrator with 20+ years of production-grade web engineering experience. He has designed, shipped, and scaled enterprise Python/PHP systems for logistics, SaaS, and public-sector clients. For the past year, he has focused exclusively on AI integrations: deploying open-source LLMs, building generative media pipelines (image, audio, video), and engineering multi-agent workflows for real production environments. His standard: reproducibility, security, cost-efficient inference—no vaporware. He documents and evaluates emerging AI tooling, separating verified capabilities from marketing noise. Technical editor at: muza-ai.eu, ai-verslas.lt, ai-naujinos.lt Connect on LinkedIn