Claude 4

Claude 4 was the release that pushed Claude Code into the mainstream. Before May 22, 2025, the tool was a side project for early adopters poking at what AI coding could actually do. After Anthropic shipped Sonnet 4 and Opus 4 that day, it turned into the thing working engineers used to ship code.

Key Specs

The jump in coding ability was a generational one. Old Claude Code had promise but was mostly an experiment. With the 4 family, it became a tool that developers reached for day in, day out.

Professional software engineering. Real codebases stopped being a problem. Multi-file refactors, debugging that ran through dependency chains, and substantive code review all became things you could trust the model with, not just toy snippets.

Agentic reliability. Both models stuck to long, multi-step instructions without wandering. Workflows that fire tool calls in sequence to reach a goal stopped falling apart halfway through and started being production-safe.

Better instruction following. Earlier releases used to skip parts of a prompt or slip in changes no one asked for. The 4 family held the line on constraints and specs much tighter.

Extended thinking arrived with Claude 3.7 Sonnet and was a real step forward. Claude 4 took that groundwork and delivered on it everywhere:

• Code quality stayed more consistent over long sessions
• Fewer made-up imports or calls to APIs that did not exist
• Sharper grasp of a project's layout and house conventions
• Gains on benchmarks pulled from actual engineering work, like SWE-bench

The jump read less like a version bump. It read like watching a junior dev step up to mid-level. You could hand it a goal and it would understand the thing you wanted, not just echo the literal words you typed.

Sonnet 4 was the daily driver. Fast enough to keep up with an interactive session. Strong enough for most of the work you threw at it. Cheap enough to run all day. Most people on Claude Code left it set as the default.

Opus 4 handled the heavy lifting. Save it for tricky architecture calls. Deep debugging runs. Any task that needed every ounce of reasoning you could throw at it. Slower and pricier, but noticeably stronger when the problem required holding a lot of moving pieces in mind at once.

Sonnet 4 is still on the API under its model ID. Opus 4 is no longer the top of the line. Opus 4.1 took that slot and tightened its production reliability.

Most users today should start on the 4.5 generation. Claude 4 still earns its spot on the timeline. The 4 family proved that coding with AI could hold up as a real engineering workflow rather than a toy.

Related Pages

• All Claude Models for the full version timeline
• Opus 4.1 for the step that replaced Opus 4
• Model selection guide for deciding what to run today