Anthropic Launches Claude Opus 4 and Sonnet 4 With Major Coding Gains

Anthropic released Claude Opus 4 and Claude Sonnet 4 this month, and the company is calling Opus 4 its most capable coding model so far. The launch happened on May 22, 2026, and it sets a new bar for how AI agents handle long, complex tasks across software engineering and research workflows.

I have been tracking the AI race closely, and this release feels different from the usual model bumps we have seen over the past year. Opus 4 is built for sustained, multi-step work. Sonnet 4 brings most of those gains to a faster, cheaper tier that developers can use at scale.

If you follow how AI skills are now the fastest route to career growth, this launch matters. It changes what entry-level coders and senior engineers can expect from their tools. Readers who track skills-based hiring trends will see why employers are paying close attention.

What Opus 4 and Sonnet 4 Actually Do

Opus 4 scored 72.5 percent on SWE-bench, the standard benchmark for real software engineering tasks, according to Anthropic’s own announcement. Sonnet 4 came in at 72.7 percent, which is striking because Sonnet runs at a fraction of the cost.

Both models can run for hours on a single task without losing track of context. Anthropic says one customer ran Opus 4 on a coding job for seven straight hours. That kind of stamina was not possible in earlier versions.

The models also include a new feature called extended thinking with tool use. The AI can pause, search the web, run code, and then come back to the original task. This is the part that excites engineers I have spoken with.

Why Coders Are Paying Attention

GitHub announced it will use Sonnet 4 as the model behind a new coding agent inside GitHub Copilot, Reuters reported. Cursor, Replit, and Rakuten also confirmed they are integrating the new models.

The pricing stays the same as Claude 3 Opus and Sonnet. That is unusual. Most AI labs raise prices when capability jumps. Anthropic kept Opus 4 at $15 per million input tokens and Sonnet 4 at $3 per million.

For context, OpenAI’s GPT-4.1 launched in April at similar price points but with weaker coding scores on independent tests, according to CNBC’s coverage of the AI model race.

What Changed Under the Hood

Anthropic trained both models with a focus on agentic behavior. That means the AI can plan, take actions, check its work, and adjust without a human prompting each step. Memory is also improved. Opus 4 can write notes to files during long tasks and read them later.

Safety testing came with the launch. Anthropic deployed Opus 4 under what it calls ASL-3 standards, a stricter tier than earlier models. The company said this was a precaution given the model’s capabilities, not a response to a specific risk.

What This Means for Workers

I think the biggest shift is for mid-level engineers. Opus 4 can handle tasks that used to take a full day in a few hours. That does not eliminate the job. It changes what the job looks like.

Junior developers should learn to direct these tools well. Senior developers should focus on architecture and review. The middle is where the pressure builds.

Companies that move fast on integration will pull ahead. Companies that wait six months will be behind by a year in productivity terms.

Bottom Line

Claude Opus 4 and Sonnet 4 are not just incremental updates. They push what AI agents can do in real work settings. Whether that is good news depends on where you sit in the labor market. For now, the models are live, the prices held steady, and the coding world has a new benchmark to chase.