Anthropic Releases Claude 4: Opus 4 and Sonnet 4 Debut Together

Anthropic released the Claude 4 series on May 22, 2025 at its first Code with Claude developer conference: the flagship Opus 4 and the cost-effective Sonnet 4, both hybrid-reasoning models.

Claude 4 is built for hours-long autonomous agentic work: at launch Anthropic demonstrated Opus 4 working continuously for hours on complex coding tasks, and media dubbed it the strongest coding model of its time. The models also improved clearly on instruction following and hallucination reduction.

Anthropic stressed that Claude 4 continues its AI safety and alignment research.

The Strategic Bet: All-In on Coding

While rivals raced on multimodality and consumer experience, Anthropic staged Claude 4's debut at a developer conference literally named Code with Claude. Claude Code went GA the same day, packaging the model into terminal workflows, and the company's MCP protocol had already become the de facto standard for tool connections (see our MCP ecosystem coverage) — model, tooling and protocol landing together completed a vertically integrated position in the coding lane.

Long-Horizon Autonomy: The New Competitive Axis

By putting "works continuously for hours" in the launch headline, Claude 4 marked the shift from benchmark scores to task completion as the yardstick: can a model decompose goals, call tools and self-correct for half a day unattended? Anthropic's own iterations kept stretching that curve — Sonnet 4.5 later claimed 30+ hours of autonomy, and Opus 4.5 pushed further. Long-horizon autonomy has been an industry-recognized axis of competition ever since Claude 4.

The Commercial Verdict

The payoff was visible in the numbers: Anthropic's run-rate revenue grew from about $1 billion in early 2025 to over $5 billion by August, driven by enterprise APIs and Claude Code, and September's Series F valued the company at $183 billion (see our coverage); Cursor, Windsurf and other mainstream tools defaulted to Claude-family models for long stretches. Coding delivered the hardest enterprise revenue of this AI cycle.

Our Take

Claude 4 is the textbook case of depth beating breadth: trading some consumer mindshare for the scenario with the strongest developer loyalty and willingness to pay. With Opus 4.1, Sonnet 4.5 and Opus 4.5 following at a quarterly cadence, the "best at coding" label got nailed to Anthropic — and the prize on that label is pricing power in AI's most commercially certain lane.