2026 年 ChatGPT vs Claude vs Gemini：33 題 3 個贏家

How we ran the test

We picked 33 real prompts drawn from common knowledge-worker workflows — coding, long-form writing, factual research, math/reasoning, structured data extraction, and multimodal tasks (image + text). Each prompt was submitted to all three models:

• GPT-4o (via ChatGPT Plus, $20/mo)
• Claude 3.5 Sonnet (via Claude Pro, $20/mo)
• Gemini 1.5 Pro with Search (via Gemini Advanced, $20/mo)

Three reviewers independently scored outputs 1-5 on:

1. Correctness — does the answer actually solve the task?
2. Reasoning — does the model explain its work and surface trade-offs?
3. Conciseness — is the response appropriate for the prompt?
4. Style — is the writing appropriate for the audience?

The total scores below are the sum across all 33 prompts (max 165 per model).

Top-line scores

Model	Coding	Writing	Research	Math	Multimodal	Total
GPT-4o	142	138	145	130	152	707
Claude 3.5 Sonnet	156	155	144	134	130	719
Gemini 1.5 Pro (Search)	128	130	158	118	145	679

Headline: Claude 3.5 Sonnet won the top-line by a small margin, driven by its dominance on coding and writing. GPT-4o led on multimodal. Gemini dominated research where Search is enabled.

By category

Coding (33 prompts)

The coding prompts ranged from “fix this bug” to “rewrite this 200-line file” to “implement this small feature from a spec”.

Model	Wins	Notes
Claude 3.5 Sonnet	22 / 33	Better at multi-file edits and matching existing code style
GPT-4o	9 / 33	Faster on single-shot completions, weaker on legacy stacks
Gemini 1.5 Pro	2 / 33	Solid at small tasks, struggles with context-tracking

Winner: Claude. By a wide margin. This is the single biggest gap in the test.

Long-form writing (10 prompts)

Drafting emails, blog posts, reports, and one short story.

Model	Wins	Notes
Claude 3.5 Sonnet	7 / 10	More natural voice, less “AI-sounding”
GPT-4o	3 / 10	Solid on shorter pieces, gets repetitive on long drafts

Winner: Claude — but only modestly.

Research with citations (10 prompts)

Fact-finding questions where a correct answer requires citing sources.

Model	Wins	Notes
Gemini 1.5 Pro with Search	8 / 10	Citations inline, up-to-date sources
GPT-4o	2 / 10	Web Browsing enabled, but citations are less clean
Claude 3.5 Sonnet	0 / 10	No web-search toggle on Pro tier

Winner: Gemini 1.5 Pro with Search. (Perplexity Pro would also be very competitive here, see our best chat assistants guide for context.)

Math and reasoning (5 prompts)

Logic puzzles, word problems, and a small bit of competition math.

Model	Wins	Notes
GPT-4o	3 / 5	Marginal lead on competition-style math
Claude 3.5 Sonnet	2 / 5	Slightly better at logic-puzzle phrasing

Winner: Roughly a tie. Both are excellent for non-research-level math.

Multimodal (image + text) (5 prompts)

Reading a screenshot, parsing a chart, describing a photo.

Model	Wins	Notes
GPT-4o	4 / 5	Best overall vision capability
Gemini 1.5 Pro	1 / 5	Solid at charts, slightly behind on photos

Winner: GPT-4o.

Pricing and free tier notes

All three converge on $20/month for the paid tier. The free tier differs sharply:

Model	Free tier	Paid tier
ChatGPT Plus	GPT-4o mini, low rate limits	$20/mo for GPT-4o + DALL·E 3
Claude Pro	Sonnet 4.5 (smaller), low rate limits	$20/mo for Claude 3.5 Sonnet + larger context
Gemini Advanced	Gemini 1.5 Flash, low rate limits	$20/mo for Gemini 1.5 Pro + 2 TB Google One

For free-tier value, Gemini Advanced gives the most — Google’s free tier has the highest rate limits and the most generous features. For paid tier value, Claude Pro or ChatGPT Plus are roughly tied depending on your task.

Privacy

All three paid tiers let you turn off data-for-training. If your work is sensitive, you should:

• ChatGPT Plus: Settings → Data Controls → disable “Improve the model for everyone”.
• Claude Pro: Settings → Privacy → disable “Help improve Claude”.
• Gemini Advanced: Google account → Gemini Apps Activity → turn off.

For maximum privacy regardless of provider, self-host Llama 2 via Ollama — your data never leaves your machine.

Final recommendation

Pick one of the three for a paid subscription, based on your dominant task:

• Coding, long-form writing, multi-file edits → Claude Pro.
• Multimodal, vision tasks, image generation, broad tool use → ChatGPT Plus.
• Search-grounded research, Google Workspace integration, free-tier value → Gemini Advanced.

If you can only subscribe to one and your work is mostly text-based, Claude Pro edges out by a hair. If your work is broad — coding one day, image editing the next, research the third — ChatGPT Plus has the best ecosystem.