中

Claude Opus 4.7 Tops Programming Leaderboard: 13% Coding Boost, But Anthropic's Ambitions Go Further

Anthropic, Claude Opus 4.7, Programming, Code Model —

Claude Opus 4.7 officially launched on April 16, and Anthropic has seriously pushed coding capabilities to new heights this time.

Honestly, when I saw the notification, I was using Claude Code to write a web scraper. Habitually clicked the update prompt and saw the model ID changed to claude-opus-4-7—my first thought was “Again?” They just updated to 4.6 last month. This iteration speed is insane.

Where’s the improvement?

The official numbers speak for themselves: SWE-bench Pro jumped from 56.8% to 64.3%, CursorBench from 62% to 70%. Just numbers? Not quite. I personally tested two tasks that previously stumped me—writing unit tests for legacy code and handling multi-file refactoring.

The results were interesting. 4.6 would “pretend to understand” these tasks, giving code that looked right but wouldn’t run. 4.7 stops and asks: “I haven’t seen this logic before, can you explain it?”

This “knowing what it doesn’t know” ability is more valuable in coding scenarios than any flashy feature.

Same price, but the landscape changed

Still $5 per million tokens input, $15 output—same as 4.6. But OpenAI’s GPT-5.4 Pro pricing is clearly more aggressive—cheaper, yes, but the capability gap is widening.

One detail many missed: Opus 4.7’s “low thinking tier” is now equivalent to 4.6’s “medium thinking tier.” What does this mean? You get better results with faster mode for the same money. Anthropic’s move is clever—no price hike, but better value.

The bigger picture

Released the same day was Claude Mythos Preview, a model specialized for cybersecurity. Combined with the earlier Enterprise Coworker, Anthropic’s product matrix is becoming clearer: general flagship (Opus), vertical specialist (Mythos), enterprise workflow (Coworker).

I sometimes wonder if Anthropic is playing a longer game? While OpenAI pushes AGI narratives and Google focuses on ecosystem integration, Anthropic seems to be taking a third path—going deep into every vertical scenario.

Slower, but more solid.

What do you think? Will Opus 4.7 become your primary coding model?

Related Posts