The short answer: there's no single winner, so pick by task. Claude leads on coding and is tied for first on writing. Gemini has the cheapest paid plan, ties Claude for the biggest context window at the $20 tier, and is tied for first on writing. ChatGPT has the widest set of tools in one app and leads on hard math with its top model. All three cost about $20 a month for the main paid plan.
Every price, plan detail and ranking below was checked on October 30, 2026, on each company's own pages and on the live Arena leaderboards. This field changes monthly, so we date everything.
The short answer: pick by task
| If you mainly want… | Best fit | Why | Watch out for |
|---|---|---|---|
| Coding and building web pages | Claude | Claude Opus 5.5 is #1 on Arena's WebDev leaderboard (1,810), ahead of GPT-6 Astra (1,773) | Claude's usage limits are shared with Claude Code |
| Writing: emails, posts, long-form | Claude or Gemini | Statistically tied in Arena's creative-writing votes: Opus 5.5 at 1,520, Gemini 4 Argon at 1,517 | Try your own prompt in both free plans |
| Math and hard reasoning | ChatGPT (top plans) or Claude | GPT-6 Astra scores 98% and Opus 5.5 95% on Epoch's hardest math set | Astra's deepest reasoning is a Pro feature ($100+) |
| Very long documents | Gemini AI Pro or Claude paid plans | Both offer windows of about 1 million tokens | ChatGPT Plus is 54K tokens, or 256K in reasoning mode |
| Images and video | Gemini or ChatGPT | Both generate images; Gemini adds video from its $4.99 plan | Anthropic's plans don't list image generation |
| Everything in one app | ChatGPT | Projects, scheduled tasks, agents (ChatGPT Work), coding (Codex), voice, apps | — |
| The lowest price | Gemini AI Plus ($4.99) | Cheapest paid plan of the three | It loses the Pro model on November 13, according to Google's email to subscribers (9to5Google) |
| Gmail and Google Docs | Gemini | Gmail from AI Plus, Docs from AI Pro | — |
What changed in October 2026
Most comparison pages you'll find still describe GPT-4o, Claude 3.5 or Gemini 1.5. This month alone:
- GPT-6 reached ChatGPT's everyday chat in early October, after the models launched in September. OpenAI's pricing page now lists GPT-6 Luna on the Free and Go plans, and GPT-6 Sol, GPT-6.1 Sol and GPT-6 Astra on Plus (OpenAI).
- Gemini's free plan moved to Flash-Lite on October 9. Its Auto mode sends most prompts to Flash-Lite and harder ones to Flash or Pro (Google).
- Gemini 4 Carbon took #1 on Arena's text leaderboard in late October, with a score Arena marks as preliminary. It isn't available to the public yet (TestingCatalog).
- Gemini's API gets more expensive: Gemini 4 Argon doubles in price on December 1, and Gemini 3.8 Flash on January 1 (Google). This affects businesses paying per use, not the $20 plans.
Plans and prices compared
The $20 plans side by side
| ChatGPT Plus | Claude Pro | Google AI Pro | |
|---|---|---|---|
| Price | $20 a month | $20 a month, or $17 billed yearly | $19.99 a month |
| Top models included | GPT-6 Sol in chat; GPT-6.1 Sol and GPT-6 Astra also listed | Claude Opus 5.5, Sonnet 5.5, Haiku 5.5 (Fable 5.1 with usage credits) | Gemini Pro and Gemini 4 Argon |
| How much you can paste in | 54K tokens (256K in reasoning mode) | Up to 1 million tokens | 1 million tokens |
| Deep research | Yes | Yes | Yes |
| Agents and coding | ChatGPT Work, Codex | Claude Code, Claude in Chrome | Jules, Chrome auto browse |
| Image and video | Images | Not listed | Images and video (including a limited Veo 3.1 Lite trial) |
| Extras | Projects, scheduled tasks | Projects, Microsoft 365 integration | 5 TB of Google storage, Gmail and Docs integration |
Sources: ChatGPT pricing, Claude pricing, Google AI plans and Claude context windows.
Free plans: what you actually get
- ChatGPT Free runs GPT-6 Luna, with what OpenAI calls unlimited everyday text chats (subject to abuse guardrails) but limited uploads, image creation, deep research and memory. Its context window is 27K tokens.
- Claude Free offers Sonnet and Haiku models (Anthropic doesn't say which versions), web search, memory and up to 5 projects, but not Research or Claude Code.
- Gemini's free plan mostly runs Flash-Lite since October 9, with a 32K context window. It includes image generation, Deep Research and Canvas.
Cheaper and pricier plans
- Cheaper: Google AI Plus is $4.99 a month (Flash-Lite and Flash, a 128K window, video generation). ChatGPT Go is $8 a month and may include ads. Claude has no plan between Free and Pro.
- Pricier: ChatGPT Pro starts at $100 a month in three usage tiers and adds Pro reasoning on GPT-6 Astra. Claude Max is $100 (5× Pro's usage) or $200 (20×). Google AI Ultra is $99.99 or $199.99.
Why nobody tells you how many messages you get
None of the three publishes a fixed message count for most models. Claude's session limits reset every five hours (Anthropic), and Gemini's refresh every five hours up to a weekly limit (Google); on both, long chats, big files and heavy features use up your allowance faster. OpenAI lists everyday text chats as unlimited (subject to abuse guardrails) and describes uploads, deep research and other tools only as "limited" or "expanded." In practice, shorter prompts and fresh chats stretch any plan further.
The leaderboard model isn't always the model you get
Benchmark headlines usually test each company's best model at its highest reasoning setting. What you chat with depends on your plan:
- ChatGPT: Free and Go get GPT-6 Luna. Plus chats with GPT-6 Sol; OpenAI also lists GPT-6.1 Sol and GPT-6 Astra on Plus, which reports say are available in ChatGPT Work and Codex rather than everyday chat. Pro gets "expanded" access and Pro reasoning powered by GPT-6 Astra.
- Claude: paid plans get Opus 5.5, Sonnet 5.5 and Haiku 5.5. Fable 5.1, Anthropic's highest-priced model, uses extra usage credits on Pro.
- Gemini: free users mostly get Flash-Lite. Google lists Gemini 4 Argon access on AI Pro and the most access on Ultra.
So "Gemini is #2 on the leaderboard" doesn't help a free Gemini user, who is mostly talking to a much smaller model.
Which AI is best for…
Writing
It's a tie between Claude and Gemini. In Arena's creative-writing category, where people vote blind on which answer they prefer, Claude Opus 5.5 scores 1,520 and Gemini 4 Argon 1,517, well within each other's margin of error (±13). No GPT-6 model is in that category's top 10 (Arena, October 27, 2026). Style is personal, so the best test is your own: give both free plans the same prompt.
Coding
Claude. Claude Opus 5.5 leads Arena's WebDev leaderboard at 1,810, followed by GPT-6 Astra (1,773), Claude Sonnet 5.5 (1,770) and Gemini 4 Carbon (1,768). Gemini 4 Argon is 11th (Arena, October 27, 2026). Claude Pro also includes Claude Code.
Everyday questions and general chat
Gemini and Claude, narrowly. On Arena's overall text leaderboard, the top publicly available models are Gemini 4 Argon (1,520), Claude Opus 5.5 (1,509) and Claude Fable 5.1 (1,504). OpenAI's highest-ranked model, GPT-5.6 Sol, is 22nd at 1,485, and GPT-6.1 Sol is 27th at 1,481 (Arena, October 27, 2026). The gaps between the top models are small, and every one of them handles everyday questions well.
Math and hard reasoning
ChatGPT's top model, with Claude close behind. On FrontierMath Tier 4, Epoch AI's hardest math set, GPT-6 Astra scores 98%, Claude Opus 5.5 95% and Gemini 4 Argon 78% (Epoch AI).
Research and fact-checking
Check the sources whichever you use. On Artificial Analysis's knowledge test, Gemini 4 Argon had a hallucination rate of 15%, the lowest of any top-scoring model it tested, against 51% for GPT-6 Astra, partly because it says it doesn't know more often (Artificial Analysis). All three can search the web and cite sources; open the citations before you rely on an answer.
Long documents and PDFs
Gemini AI Pro or a paid Claude plan. Both handle about a million tokens, enough for several long reports at once. ChatGPT Plus is limited to 54K tokens in normal mode and 256K in reasoning mode, which is the smallest of the three $20 plans.
Images and video
Gemini or ChatGPT. Both generate images, and Gemini adds video generation from its $4.99 AI Plus plan up. Anthropic's plan pages don't list image or video generation for Claude.
Speed
The setting matters more than the brand. Every app has fast modes that answer in seconds and reasoning modes that think first. Deep reasoning on a top model can take minutes. Use the fast mode for quick questions and save reasoning for hard problems.
ChatGPT vs Claude
Choose ChatGPT for the broadest toolset in one app (agents, Codex, scheduled tasks, image generation, voice) and for top-end math. Choose Claude for coding, careful long-form writing and much longer documents: a paid Claude plan accepts up to a million tokens against ChatGPT Plus's 54K or 256K. Both cost $20 a month; Claude is $17 a month if you pay yearly.
Claude vs Gemini
These two are closest on quality: tied for writing, and the top two publicly available families on Arena's overall text leaderboard. Claude pulls ahead on coding. Gemini wins on price (from $4.99), Google integration, image and video generation, and extras like 5 TB of storage on AI Pro.
ChatGPT vs Gemini
ChatGPT has more tools and stronger top-end math. Gemini scores higher on general chat and writing votes, is cheaper at the low end, holds far longer documents on its $20 plan, and works inside Gmail and Docs.
Privacy: who trains on your chats, and how to turn it off
All three use personal-plan chats to improve their models unless you opt out, and all three have a mode that isn't used for training.
| App | Default | How to opt out | Private mode |
|---|---|---|---|
| ChatGPT | Chats can be used for training | Settings → Data Controls → turn off "Improve the model for everyone" | Temporary Chat, deleted within 30 days |
| Claude | Used for training unless you turn it off | Settings → Privacy → turn off "Help improve our AI models" | Incognito chats, never used for training |
| Gemini | Keep Activity is on by default for users 18 and over | Settings & help → Activity → Turn off | Temporary chats, kept for 72 hours |
Google also says human reviewers read a sample of Gemini chats, after disconnecting them from your account, and keeps reviewed chats for up to three years even if you delete your activity. Business plans from OpenAI and Anthropic don't train on your data by default (OpenAI, Anthropic). Sources for the table: OpenAI, Anthropic, Google and Google's privacy hub.
For businesses on the API: the cheapest model that does the job
If you pay per use through the API, the choice is different: each company sells models at very different prices, and most jobs don't need the top one. PromptSaver sorts models into four levels of its own; here's the cheapest model at each level today:
| Task level | Anthropic | OpenAI | |
|---|---|---|---|
| Light: short copy, rewrites, tagging | Claude Haiku 5.5 ($0.10 / $0.50; more above 100K input tokens) | GPT-6 Luna ($0.10 / $0.50) | Gemini 3.1 Flash-Lite ($0.25 / $1.50) |
| Everyday: emails, posts, summaries | Claude Sonnet 5.5 ($2 / $10) | GPT-6.1 Terra ($1 / $5) | Gemini 3.8 Flash ($0.75 / $3.75) |
| Hard reasoning | Claude Opus 5.5 ($4 / $20) | GPT-6.1 Sol ($2 / $10) | Gemini 4 Argon ($2 / $10 until Nov 30) |
| Frontier | Claude Fable 5.1 ($10 / $50) | GPT-6 Astra ($10 / $50) | No model at this level yet |
Prices are per million input / output tokens. OpenAI's GPT-6 models also charge more once a request's input passes 272K tokens. By our estimate, Claude counts roughly 35% more tokens than ChatGPT for the same English text, so its real cost runs higher at the same list price. See the full AI pricing comparison, or run your own numbers in the AI cost calculator.
Should you pay for two?
Many heavy users do. Claude Pro plus Google AI Plus costs about $25 a month and covers coding, writing, long documents, images and video. If you only pay for one, choose by your main task in the table at the top, and use the other two companies' free plans for second opinions.
Frequently asked questions
Which is better in 2026: ChatGPT, Claude or Gemini?
None wins at everything. As of late October 2026, Claude leads coding, Claude and Gemini are tied for writing and close at the top for general chat, and ChatGPT's top model leads hard math. Pick the one that's best at what you do most.
Is ChatGPT Plus, Claude Pro or Google AI Pro worth $20?
All three are about $20 a month, and Claude Pro is $17 a month billed yearly. Google AI Pro adds 5 TB of storage and, like Claude Pro, a context window of about a million tokens. Claude Pro includes Claude Code. ChatGPT Plus includes the widest set of tools, such as agents, Codex and scheduled tasks.
Which AI is best for coding?
Claude. Claude Opus 5.5 is #1 on Arena's WebDev leaderboard at 1,810, with GPT-6 Astra second at 1,773, as of October 27, 2026.
Which AI is best for writing?
It's a tie between Claude and Gemini. In Arena's blind creative-writing votes, Claude Opus 5.5 scores 1,520 and Gemini 4 Argon 1,517, within the margin of error. Try your own prompt in both.
Which model does ChatGPT use now?
According to OpenAI's pricing page, Free and Go use GPT-6 Luna, and Plus adds GPT-6 Sol, GPT-6.1 Sol and GPT-6 Astra (reports say the last two are in ChatGPT Work and Codex rather than everyday chat). Pro gets expanded access plus Pro reasoning powered by GPT-6 Astra.
Which AI can handle the longest documents?
Gemini AI Pro and Ultra and Claude's paid plans accept about 1 million tokens. ChatGPT Plus accepts 54K tokens in normal mode and 256K in reasoning mode, and Pro 128K and 400K.
Do ChatGPT, Claude and Gemini train on my chats?
On personal plans, all three can use your chats unless you opt out. Turn off "Improve the model for everyone" in ChatGPT, "Help improve our AI models" in Claude, or Keep Activity in Gemini.
What's the cheapest paid AI plan?
Google AI Plus at $4.99 a month. ChatGPT Go is $8 a month and may include ads. Claude's cheapest paid plan is Pro at $17 to $20 a month.
Can Claude make images?
Anthropic's plan pages don't list image or video generation. ChatGPT generates images on every plan, and Gemini generates images for free and video from AI Plus up.
How we checked this
We read each company's pricing, plan and privacy pages on October 30, 2026, and the live Arena leaderboards (text, creative writing and WebDev, all dated October 27, 2026). Benchmark scores from Epoch AI and Artificial Analysis are dated on their pages. Arena scores come from blind votes by real users, and top scores are often within a few points of each other, so treat close rankings as ties. PromptSaver isn't affiliated with OpenAI, Anthropic or Google, and nobody paid to be included.