Anthropic's July 2026 update to Claude 3.5 Sonnet has shifted the competitive landscape for enterprise coding tools. The model now scores 73.4% on SWE-bench Verified — the benchmark that measures an AI's ability to resolve real GitHub issues in open-source software repositories — placing it ahead of GPT-4o (68.1%) and within 4 percentage points of OpenAI's reasoning-focused o3 model.
The improvement is not evenly distributed across task types. Claude 3.5 Sonnet shows the largest gains in multi-file refactoring, test generation, and debugging tasks that require holding large amounts of context while making targeted changes. On single-function implementation tasks, the gap between models is smaller. This profile makes it particularly useful for enterprise engineering teams working with large codebases rather than greenfield development.
On mathematical reasoning, Claude 3.5 Sonnet now scores 81.2% on MATH-500, up from 74.3% in the previous version. This narrows the gap with o3 (89.4%) but does not close it — OpenAI's reasoning model maintains a lead on problems requiring multi-step formal proof and olympiad-level competition mathematics.
For developers using Anthropic's API, the practical impact is visible in token efficiency. The updated model achieves the same task completion with roughly 15% fewer output tokens on average for coding tasks, reducing costs for high-volume deployments. The 200,000-token context window remains unchanged.
Enterprise adoption is accelerating. Anthropic disclosed that API revenue grew 4x year-over-year in the first half of 2026, driven primarily by engineering tool integrations at software companies. The Cursor AI coding editor, which offers Claude 3.5 Sonnet as one of its model options, reports that it is the most-selected model among users who pay for the Pro tier.
Anthropic's roadmap points toward Claude 4 in late 2026, which is expected to add native tool use improvements and better performance on agentic multi-step tasks — the frontier where all major labs are now competing most aggressively.