Claude 4.7 vs Gemini 2.5 Pro: coding, writing, and vision benchmarks
Reproducible head-to-head — fifteen prompts you can run yourself, scored on first-pass quality and how much correction each model needed.
Claude 4.7 and Gemini 2.5 Pro are the two most-asked-about alternatives to GPT-5 in mid-2026. They occupy similar ground — flagship reasoning, long-ish context, vision input — but they trade wins on specific tasks. We ran fifteen reproducible prompts (links to the public gist at the end) across coding, writing, and vision, scored each on first-pass quality and how many follow-up turns it took to get a usable answer. Here's where each one actually wins.
Coding (5 prompts)
| Task | Claude 4.7 | Gemini 2.5 Pro |
|---|---|---|
| Write a Python parser for X format | ✓ first try | ✓ first try |
| Refactor 300-line module | ✓ first try | needed 1 correction |
| Debug an async race condition | needed 2 corrections | needed 1 correction |
| Add tests for legacy code | ✓ first try | ✓ first try |
| Translate Python → TypeScript | needed 1 correction | needed 2 corrections |
Claude won 3, Gemini won 1, one tie. Where Gemini lost on coding was refactoring — it tended to introduce stylistic changes (different bracket placement, different variable naming) along with the requested refactor. Claude kept the existing style more faithfully.
Where Gemini WON was the async debugging — it identified the race condition faster and proposed a cleaner fix.
Writing (5 prompts)
| Task | Claude 4.7 | Gemini 2.5 Pro |
|---|---|---|
| 800-word essay (technical topic) | ✓ first try | needed 1 correction |
| Rewrite for a different audience | ✓ first try | ✓ first try |
| Headline + dek for marketing | needed 1 correction | ✓ first try |
| Summarize a 30-page report | ✓ first try | needed 2 corrections |
| Edit for clarity (line edits) | ✓ first try | needed 1 correction |
Claude won 4, Gemini won 1. Claude is more comfortable in essay mode (longer-form, opinionated, structured argument). Gemini felt stronger on punchy short-form (marketing headlines, social copy).
Vision (5 prompts)
| Task | Claude 4.7 | Gemini 2.5 Pro |
|---|---|---|
| Describe screenshot of a UI | ✓ | ✓ (more detail) |
| Extract data from a table image | needed 1 correction | ✓ first try |
| Explain a hand-drawn diagram | ✓ first try | needed 1 correction |
| Identify objects in a photo | ✓ | ✓ |
| OCR + correct typos in scanned page | needed 1 correction | ✓ first try |
Tied at 5-5 if you count "needed 1 correction" as half a win, but the split is structural: Gemini wins on text-in-image tasks (OCR, tables), Claude wins on diagram interpretation (especially hand-drawn).
The honest summary
For most readers, the choice is the same as in our broader comparison piece: you don't have to pick one. With oran.chat, you switch per question — Claude for the refactor, Gemini for the OCR, GPT-5 for the marketing headline — without three subscriptions.
If you ARE picking one: Claude 4.7 for code-and-writing-heavy work, Gemini 2.5 Pro if your tasks include a lot of image-text extraction or you're already in Google's tooling ecosystem.
Reproducibility
The fifteen prompts are at our public gist (linked from this page in the live version). Run them yourself, score with the same rubric, and tell us where your results diverged.
More head-to-heads in Comparisons.