Claude vs Gemini.
Anthropic's assistant against Google's: the current models, what each can read and make, where each works, what the plans and the API cost, and which to pick for which job.
Claude
GeminiFeature by feature.
Models, as of November 2026
| Feature | Claude | Gemini |
|---|---|---|
| Top model | Claude Fable 5.2 | Gemini 4 Carbon |
| Recommended for most work | Claude Opus 5.5 | Gemini 4 Argon |
| On the free plan | Claude Sonnet and Haiku | Gemini Flash-Lite, Flash for harder prompts |
| Paid plans add | Claude Opus 5.5 and Fable 5.2 | Gemini Pro and Flash; Argon on AI Pro; Carbon on AI Ultra |
Reading & Making
| Feature | Claude | Gemini |
|---|---|---|
| Reads (API) | Text, images and PDFs | Text, images, audio, video and PDFs |
| Makes images | No; diagrams and charts in code | Yes, with Nano Banana 2 |
| Makes video | No | Short clips with sound, on paid plans |
| Voice | Voice mode, in beta | Gemini Live, with your camera |
| Deep research | Research, on paid plans | On every plan |
Work & Tools
| Feature | Claude | Gemini |
|---|---|---|
| Office apps | Microsoft 365, on paid plans | Gmail, Docs and Sheets, on paid plans |
| Shared assistants | Projects and Skills | Skills, which replaced Gems |
| Memory of past chats | On every plan | Personal Intelligence, optional, in beta |
API, per million tokens
| Feature | Claude | Gemini |
|---|---|---|
| Context window | 1 million tokens | 1,048,576 tokens |
| Longest reply | 128,000 tokens | 65,536 by default, more with long decode |
| Top model | $8 in, $40 out (Fable 5.2) | $2 in, $10 out to 30 November, then $4 and $20 (Gemini 4 Carbon) |
| Recommended model | $4 in, $20 out (Opus 5.5) | $2 in, $10 out to 30 November, then $4 and $20 (Gemini 4 Argon) |
| Smallest model | $0.10 in, $0.50 out (Haiku 5.5, prompts to 100,000 tokens) | $0.30 in, $2.50 out (Gemini 3.5 Flash-Lite) |
| Long prompts | Standard price to 1 million tokens, Haiku aside | One price at any length |
| Free API tier | Not listed | On Flash and Flash-Lite |
Privacy
| Feature | Claude | Gemini |
|---|---|---|
| Consumer chats used for training | Your choice, in privacy settings | Yes, while Keep Activity is on (the default) |
| Business plans | No training by default | No training without your permission |
| API | Not used for training by default | Not used on the paid tier; used on the free tier |
Plans & pricing.
| Plan | Claude | Gemini |
|---|---|---|
| Free | $0 | $0 |
| Entry plan | None between Free and Pro | Google AI Plus: $4.99/month |
| Standard plan | Pro: $20/month, or $17/month billed yearly | Google AI Pro: $19.99/month, with 5 TB storage |
| Heavy use | Max: $100/month; Ultra: $200/month | Google AI Ultra: $99.99 or $199.99/month |
| Team seat | Team: $20/seat billed yearly, $25 monthly | Gemini Enterprise Business: from $21/seat/month |
| Enterprise | $20/seat/month billed yearly plus API usage, 20 seats minimum | Gemini Enterprise Standard and Plus: from $30/seat/month |
US dollars as of November 2026, from claude.com/pricing, gemini.google's US plans page and each company's API pricing page. Prices can differ by country.
Our conclusion.
Pick Gemini when your work runs on audio, video or Google Workspace, when you want images and video made in the app, or when API cost decides it. Pick Claude when your team works in Microsoft 365, when you need long replies from the API (up to 128,000 tokens each, against Gemini's default 65,536), or when you want to choose for yourself whether your chats train its models.
Gemini reads audio and video directly, makes images and short videos, works inside Gmail, Docs and Sheets on its paid plans, and has a $4.99 plan below its $19.99 Google AI Pro. Claude makes no images or video, and its first paid plan is Pro at $20 a month, or $17 billed yearly.
Through the API, Google's top model costs $2 in and $10 out per million tokens until 30 November 2026, then $4 and $20; Anthropic's top model costs $8 and $40, and its recommended Opus 5.5 costs $4 and $20. Claude's smallest model is cheaper than Google's. On consumer plans Gemini trains on your chats while Keep Activity is on, its default, and Claude asks you to choose.
Models change often; we review this page every quarter.
Where we come in.
- AI ChatbotsA bot that answers in whichever language the visitor opened with, and knows when to fetch a person.See the page
- AI AgentsAgents scoped to a defined job with a clear handover, rather than an open-ended promise nobody can audit.See the page
- AI Business ToolsInternal tools shaped around your data and your week, rather than a generic SaaS you bend yourself around.See the page
- AIPractical automation for small teams: the repetitive work handled, so the people you have can spend their time on the work that needs them.See the page
Other comparisons.
ChatGPT vs Claude
Compare ChatGPT and Claude on models, tools, plans, API prices and privacy.
See ChatGPT vs ClaudeChatGPT vs Gemini
Compare ChatGPT and Gemini on models, tools, plans, API prices and privacy.
See ChatGPT vs Gemini
ChatGPTFAQ
Questions,
answered.
No neutral test settles it for your code. Both companies publish coding results for their own models, on their own tests, and Claude includes its Claude Code agent on paid plans. Try both on a real task from your codebase.
Gemini, at most levels. Its app plans start at $4.99 a month, and through the API its top model costs $2 in and $10 out per million tokens until 30 November 2026, then $4 and $20, against $8 and $40 for Claude's. Claude's smallest model, Haiku 5.5, costs less than Gemini's Flash-Lite.
No. Claude's current models read text and images, plus PDFs. Gemini's models also read audio and video.
No. Claude does not make photos or illustrations; it can draw diagrams and charts in code. Gemini makes images with Nano Banana 2 and short videos on paid plans.
They are about the same in the API: 1 million tokens for Claude and 1,048,576 for Gemini. In the Gemini app the window is 1 million tokens on Google AI Pro and Ultra, 128,000 on Plus and 32,000 without a plan.
Anthropic and Google each publish results for their own models, mostly on different tests run their own way, so the numbers do not line up well enough to rank the two. Each company's own model pages give its results.
