Gemini 2.5 with 1 million tokens: 4 things that work, one that doesn't
Google extended Gemini Advanced to a million-token context window and added Google Drive code uploads, but the competition still holds a clear lead in one area.
Is Gemini 2.5 Pro really better than Claude Sonnet 4.6 for serious work? It depends on what you do. In May, Google extended Gemini Advanced to a million tokens of context, added file uploads from Google Drive, and improved document analysis. That means you can finally drop a 2,000-page PDF into a single prompt and ask the model what it thinks of chapter thirteen.
We tested the new version for a week in real use, and here is where Gemini now works better, and where it still trails the competition.
1. Code uploads straight from Google Drive (works)
The most practical new feature. You can open a folder of code in Google Drive and upload several files at once. Gemini understands the project structure, not just an individual file. Supported formats include C, Python, SQL and most standard programming languages.
Compared to Anthropic’s Claude Sonnet, where you have to paste file by file or use the Projects feature, this is noticeably faster. Over a half-hour work session, the saving is around ten minutes.
2. Long-document summarisation (works very well)
We uploaded a 340-page legal-regulation PDF and asked Gemini to pull the key arguments by chapter. The output came back in about 40 seconds, with concrete page numbers for each argument, something most other models miss.
The million-token version has the capacity to hold the entire document in mind without slicing it. For lawyers, journalists, researchers, this is a genuine everyday tool.
3. Workspace subscription access (works)
Google made Gemini 2.5 Pro available to users of Google Workspace Business and Education plans. If your company already pays for Workspace, you get access at no extra cost. That is why so many teams already using Gmail for work will naturally move to Gemini instead of paying separately for ChatGPT Plus.
4. Native reasoning (works, with a caveat)
Gemini 2.5 Pro has built-in so-called thinking, where the model runs an internal analysis step before answering. It shows up on math and logic tasks, where the result is noticeably better than on the fast version. For writing, where complex reasoning is not required, the difference is smaller.
What still doesn’t work, creative writing
This is where Gemini still trails Claude Sonnet and even ChatGPT. The text it generates has a recognisable tone you can identify as Google: formal, predictable, full of phrases that read like they have been through a corporate filter.
For a journalistic piece or a creative format, the model needs a lot of editing. Anthropic still holds the lead here, especially when you are asking for a specific author’s voice.
Differences table
| Feature | Gemini 2.5 Pro | Claude Sonnet 4.6 |
|---|---|---|
| Context | 1,000,000 tokens | 1,000,000 tokens (beta) |
| Drive upload | Yes, native | No |
| Workspace integration | Built-in | Through third-party tools |
| Writing in author’s voice | Weaker | Stronger |
| Subscription price | $20 / month | $20 / month |
Our recommendation
If you use Google Workspace for work and do a lot with documents and code, Gemini 2.5 Pro is now a serious option. If you write for a living, stay on Claude or use a combination. The subscription is the same, twenty dollars a month, and the two tools do not really conflict.
Privacy and data
An important detail many users miss: under Google’s terms, data you upload through the Gemini Workspace plan is not used to train the model. The same is true at Anthropic. If you use the free version, both vendors reserve the right to use your conversations to improve the model, with an option to turn that off in settings.
For companies handling sensitive data (lawyers, doctors, accountants), a Workspace plan is the only responsible choice.
The next major Gemini version is expected at Google I/O in 2027.
