Gemini 2.5 just leapfrogged Claude 3.7 Sonnet at coding — but nobody noticed

A couple days ago Google announced the release of Gemini 2.5, its latest AI model.

To call it "impressive" is an understatement. Early benchmarks suggest it codes even better than Claude 3.7 Sonnet, which has held the crown as the top-performing AI model for developers for the past few months:

Given all the hype around vibe coding lately, you'd expect Google's new model to be dominating the online conversation.

And yet? Few people are really talking about it.

The first and most obvious reason for this is that OpenAI's new image gen model for GPT-4o has completely eclipsed Google in the news cycle:

But the other reason is something Google has more control over: Gemini 2.5 just isn't that easy to use. Which comes down to two factors:

• The $20/month price tag

• Limited integration with code editors

The two barriers to usage

1. The $20/month price tag

While Google’s $20/month fee for Gemini 2.5 pales in comparison to OpenAI’s top-tier $200/month enterprise plans, it still feels prohibitive to indie developers who are often juggling multiple AI subscriptions.

Combine this with a subscription for Cursor (plus other specialized AI tools), and the costs can pile up.

2. Limited integration with Cursor

Cursor, the most popular AI-powered code editor, has become the go-to environment for vibe coding. Historically, both OpenAI’s GPT-4 and Anthropic’s Claude integrated seamlessly with Cursor.

But Gemini 2.5 is off to a rocky start.

During Gemini 2.5’s initial release window, Cursor didn't offer official support, so developers had to rely on clunky workarounds.

And even after official Cursor support rolled out, the rate limits for Gemini 2.5 remained painfully small. For example, I hit my daily usage cap after only a handful of code generations:

Workaround: Google AI Studio

For now, the best workaround available is to use Google AI Studio, which — for the time being — allows free access to Gemini 2.5's capabilities.

But even this has some annoying limitations:

• Token limit issues: Once you hit the token limit in a single session, you have to delete older chat content to keep using the model.

• No project management: Without Cursor’s project folders, file organization, or advanced debugging tools, you’ll be pasting code snippets in and out of the browser all day.

• No “Project Rules”: Cursor’s beloved Project Rules feature allows advanced settings per project (like coding style, linting rules, or security guidelines). In Google AI Studio, you’re starting from scratch every time you open a new session.

My personal workflow has been to generate code snippets in Google AI Studio, then manually copy these outputs to Cursor for organizational purposes.

Meanwhile, if Google works with Cursor to increase the rate limit issue, Gemini might take the lead in the AI race for the first time since ChatGPT brought AI to the masses: