Claude vs ChatGPT: Which AI Assistant Fits Which Job

Claude vs ChatGPT at a glance

Everything in this table comes from the two vendors' own pages, read on 20 October 2026, the same day as every other figure on this page. Both are general assistants built on large language models, and both run on web, desktop and phone. Mindber's directory keeps a profile of each, Claude and ChatGPT, and Compare sets the two profiles side by side. Google's Gemini gets the same treatment against ChatGPT in our AI comparisons topic.

Which models do Claude and ChatGPT run?

In Claude you pick the model yourself. Most of the lineup changed in about two weeks: Opus 5.5 arrived on 22 September 2026, Sonnet 5.5 on 28 September and Haiku 5.5 on 7 October. Anthropic's pitch for Opus 5.5 is that it "performs at the level of Claude Fable 5.1 on most work" while costing 40% less to run than Opus 5. Older models such as Claude Opus 4.8 keep their own pages in Mindber's directory.

Fable 5.1 sits above them all, with strings attached. Pro users reach it only through paid usage credits, and Max users get it within 50% of their weekly limits. Free users get neither Fable nor Opus.

It has been switched off once. A US export-control directive took Fable 5 offline on 12 June 2026, as we reported at the time. Anthropic says the controls were lifted on 30 June, and access came back on 1 July.

ChatGPT keeps model names out of sight. You choose a thinking level instead, from a picker OpenAI introduced on 10 June 2026: Instant, Medium and High, plus Extra High and Pro on Pro plans. Behind it, since 7 October 2026, is GPT-6. Plus, Pro, Business and Enterprise get GPT-6 Sol, and Free and Go get GPT-6 Luna. OpenAI's help center still named GPT-5.6 Luna as the Free default while that rollout ran.

OpenAI's strongest models aren't in the chat box at all. GPT-6 Astra, its flagship since 3 September 2026, and GPT-6.1 Sol, out on 29 September, live in ChatGPT Work and Codex. OpenAI says 6.1 Sol hasn't reached Chat. Both apps offer reasoning models as well, which think before they answer and cost you time and usage for it.

Latest models side by side: score, price and speed

You don't have to take either vendor's word for how the models compare. Artificial Analysis runs its own battery of tests and publishes an Intelligence Index, what each test task cost and how fast the answer streamed. Arena ranks models from blind votes. We read both the same day and paired the models by tier. Mindber's Model Arena tracks many more models on the same two axes, capability and price.

API prices are per million tokens. "Index" is the Artificial Analysis Intelligence Index at each model's highest reasoning setting, and "a task" is its average cost to run one test from that index. Claude models run there with Anthropic's safety fallback switched on, as they do for customers.

API prices are per million tokens. "Index" is the Artificial Analysis Intelligence Index at each model's highest reasoning setting, and "a task" is its average cost to run one test from that index. Claude models run there with Anthropic's safety fallback switched on, as they do for customers.

Developers paying by the token will find the two price lists oddly matched. Sonnet 5.5 and GPT-6.1 Sol charge the same rates, and so do Haiku 5.5 and GPT-6 Luna. Fable 5.1, which isn't in the table, matches GPT-6 Astra at $10 in and $50 out. On every one of these models, output costs five times as much as input. The LLM cost calculator turns rates like these into a monthly bill, and the token-based pricing entry explains the billing model.

ClaudeOpenAI

Top: Opus 5.5 vs GPT-6 Astra

Claude

58 pts

OpenAI

53 pts

Middle: Sonnet 5.5 vs GPT-6.1 Sol

Claude

56 pts

OpenAI

52 pts

Small: Haiku 5.5 vs GPT-6 Luna

Claude

43 pts

OpenAI

38 pts

Source: Artificial Analysis LLM leaderboard, highest reasoning setting per model, Oct 20, 2026

ClaudeOpenAI

Top: Opus 5.5 vs GPT-6 Astra

Claude

$5.98

OpenAI

$3.26

Middle: Sonnet 5.5 vs GPT-6.1 Sol

Claude

$5.46

OpenAI

$0.72

Small: Haiku 5.5 vs GPT-6 Luna

Claude

$0.21

OpenAI

$0.07

Source: Artificial Analysis LLM leaderboard, cost per Intelligence Index task, Oct 20, 2026

ClaudeOpenAI

Top: Opus 5.5 vs GPT-6 Astra

Claude

95 tok/s

OpenAI

66 tok/s

Middle: Sonnet 5.5 vs GPT-6.1 Sol

Claude

134 tok/s

OpenAI

55 tok/s

Small: Haiku 5.5 vs GPT-6 Luna

Claude

232 tok/s

OpenAI

124 tok/s

Source: Artificial Analysis LLM leaderboard, median output speed at the highest setting, Oct 20, 2026

Cost is where the tiers part ways.

At the top, Opus beats Astra by five points but costs $5.98 a task to Astra's $3.26. Its token rates are lower, though. The gap comes from Opus spending far more tokens on each task. Drop both to the high setting and the bills almost meet: Opus scores 54 for $1.82, Astra 51 for $1.73. Same money, three more points for Opus.

The middle tier is OpenAI's best deal. GPT-6.1 Sol scores 52 for 72 cents a task. Sonnet reaches 56 at its highest setting, but only by spending $5.46, more than seven times as much. At the high setting Sonnet falls to 47 for 88 cents, while 6.1 Sol manages 50 for 32 cents. For big API batches where cost per task decides, 6.1 Sol is the better buy.

Speed runs the other way. Claude streams faster in all three tiers, and Sonnet does it at more than twice the rate of 6.1 Sol. Waiting for a complete 500-token answer at the high setting takes about 16 seconds with Sonnet and about 60 with 6.1 Sol. If a person is watching the reply appear, that gap matters more than the index.

Small models are for volume. GPT-6 Luna costs 7 cents a task and scores 38. Haiku costs 21 cents, scores 43 and streams almost twice as fast.

One surprise sits in Anthropic's own lineup. Fable 5.1, the top tier on Claude's plan page, scores 53 for $7.63 a task on the same leaderboard, below Opus. That fits Anthropic's claim that Opus 5.5 works at Fable level, and it makes Opus the better buy for most work.

GPT-6 Sol, which answers in ChatGPT's Chat tab on paid plans, has no Artificial Analysis score yet. It ranks ninth on Arena's WebDev board.

Writing and editing

Arena's creative-writing leaderboard is one of the few independent writing tests. Voters see two anonymous answers and pick one. Claude Opus 5.5 was first with 1,519 points. OpenAI's best was GPT-5.6 Sol, in 20th place, and 6.1 Sol was 30th.

The board ranks models reached through their APIs at fixed effort settings, not the apps you'd open, so it says more about the engines underneath than about the products. GPT-6 Sol, the model behind ChatGPT's Chat tab, isn't in the category's top thirty.

Claude now builds more of the finished document. Ask for a report or a deck and Claude Docs or Claude Slides, both in beta, produces it inside the conversation, ready to download as PowerPoint or PDF. Artifacts, the panel that keeps a draft beside the chat, comes with every plan. OpenAI went another way this autumn. Its GPT-6 update added Intelligent UI, which drops charts, forms and buttons into replies: useful for explaining something, little help with a long draft.

Either model gets closer to your voice with a brief. Name the reader, paste a sample of your style and say what format you want, as our prompt engineering guide shows step by step. Dedicated tools in the AI writing software category add brand-voice settings and shared templates on top of models like these.

Coding

On code they're close. Arena's WebDev leaderboard has voters compare web apps built from the same prompt, and it put Claude Opus 5.5 first with 1,807 points. GPT-6 Astra (1,774), Claude Sonnet 5.5 (1,768) and GPT-6.1 Sol (1,753) came next, and Arena's own rank spreads for places two to four overlap. Call those three a tie.

Each company claims an edge, and we found no independent check of either. OpenAI says 6.1 Sol matches Astra on its DeepSWE coding test at roughly one-fifth of the cost. Anthropic's announcement reports gains for Opus 5.5 over Opus 5 on coding and agent tasks.

Day to day, you'll notice the agent more than the model. Claude Code works from the terminal and the code editor, and Codex comes bundled with ChatGPT. Mindber files both under AI coding assistants, next to browser tools like Codex in Chrome. Developers who get the most from either one have the agent check its own work in a loop, the approach our loop engineering post describes.

Research, files and long documents

Claude at least tells you its ceiling. Its plan page lists a context window of up to 1M tokens in the app, depending on the model. OpenAI gives no in-app figure and only calls context limited on Free and larger on paid plans, though its API docs list 1.05M tokens for GPT-6 Astra and GPT-6.1 Sol.

Web search and file uploads work on every plan of both. Claude can run code to create and edit files, even on Free, but its Research mode starts at Pro. ChatGPT gives Free users a limited taste of deep research and Plus users more. For stored files, OpenAI lists 500 MB of Library space on Free, 20 GB of storage on Plus and 100 GB on Pro.

OpenAI claims the edge on PDFs. It says 6.1 Sol beats Opus 5.5 on its GDP.pdf test at under half the cost per task, which stays a claim until someone else runs it. Both assistants can state errors as facts, the problem known as hallucination, so quotes and figures need checking against the original file before you reuse them.

Images, voice and other media

ChatGPT makes images on every plan, with limits on Free and Go and better quality on Plus. Claude doesn't. Anthropic's help center says Claude "doesn't generate photos or illustrations the way image-generation tools do". What it will draw is a diagram, a chart or an interactive visual in HTML and SVG, a feature still in beta on web and desktop.

Make images every week? That alone may settle it. The alternative is Claude plus a separate image generator, which you can find by describing the job to Mindber Ask or browsing Discover.

Voice mode is on every plan of both, rationed on ChatGPT's Free plan. Both read images you upload, so both are multimodal on the way in, even though only one makes pictures.

Plans and prices compared

Prices below are US dollars a month before tax, from the same reading of the vendors' pages.

Our Claude Pro vs ChatGPT Plus comparison goes through the two standard plans line by line, including what each one leaves out. Neither sticker price is the whole bill, and the true cost of AI tools post lists what it misses.

Claude Max comes with $100 to $200 a month in API credits on top, real money if you build on Anthropic's platform. On ChatGPT, anyone signing up for Pro 200 now gets a smaller allowance than before. Existing subscribers keep the old one until 29 October 2026.

Usage limits

Neither side sells unlimited use of its top model. Claude meters you in sessions that reset every five hours, and Anthropic says it may add weekly and monthly caps on top. Pro users can buy usage credits to keep going past the included amount, and Claude shows what you have left under Settings, then Usage. When Opus 5.5 launched, Anthropic raised the five-hour limits on Pro, Max and Team.

ChatGPT's Free plan works the other way round. Everyday text chat is unlimited, subject to abuse safeguards, while uploads and images are capped separately. OpenAI says Plus may cap messages, mostly when demand is high. Even Pro doesn't get unlimited use of the GPT-6 Pro model.

Privacy and data controls

Since a 2025 update to its consumer terms, Anthropic asks Free, Pro and Max users whether their chats can train its models. Say yes and it keeps new chats for five years. Say no and the retention period stays at 30 days. Incognito chats are never used for training, while thumbs-up or thumbs-down feedback is stored for up to five years. None of this applies to Claude for Work, Claude for Education, Claude for Government or the API.

OpenAI puts its opt-out in ChatGPT's settings and in the OpenAI Privacy Portal. Temporary chats aren't used for training. One catch on personal plans: rating a reply can still let OpenAI use that whole conversation, even after you opt out. Business, Enterprise, Edu and API data isn't used for training by default.

Staff pasting client work into personal accounts is what's known as shadow AI, and the team plans' admin controls exist to stop it, whichever vendor you pick. Teams in Singapore can take the vendor questions from our PDPA and PDPC procurement guide to either company.

Apps, connectors and agents

Both assistants can now act for you, which makes each one an AI agent as much as a chatbot. An agent that can send email or change files is safer on read-only access for its first few days. Claude links to other apps through connectors on every plan, many of them built on the Model Context Protocol. Pro adds Claude in Chrome, Claude for Microsoft 365 and Outlook, and tasks you can hand off or schedule. Anthropic is folding Claude Cowork into the main app as well, Pro and Max first.

ChatGPT has native apps for web, iOS, Android, macOS and Windows. Its Work agent, launched on 9 July 2026, takes actions across your apps and files, with limited access on Free. Plus adds email and calendar connections. Pro adds Dots, which OpenAI calls "your always-on agents".

Frequently asked questions

Is Claude better than ChatGPT?

For writing and document work, yes. For images, no, since Claude doesn't make them. Without paying, ChatGPT does more: its free plan includes image creation, some deep research and a coding agent, and Claude Free has none of the three.

Is Claude or ChatGPT better for coding?

It's close to a tie on public rankings, so the plan you're on can decide it. Codex comes with every ChatGPT plan, Free included with limits, while Claude Code starts at Claude Pro. On a paid plan, run both agents on your own repository for a week and keep the one that suits how you work.

Which is better for writing, Claude or ChatGPT?

Claude, but judge it on the right plan. The model that leads the writing ranking, Opus 5.5, comes with Pro, while Claude Free runs Sonnet and Haiku. A free-plan trial shows you the smaller models, not the one that won.

Are Claude Pro and ChatGPT Plus the same price?

In the US, yes: $20 a month each before tax. Claude Pro drops to $17 a month if you pay for a year up front, and ChatGPT Plus has no yearly option. Anthropic says Pro pricing varies by region.

Can Claude generate images like ChatGPT?

No, on any plan, and no setting turns it on. The usual workaround is a separate image tool next to Claude. Claude can still write the prompt you paste into that tool and describe what you want changed between versions.

Do Claude and ChatGPT use my chats to train their models?

On personal plans they can, until you opt out. In Claude, change the model-training choice in Privacy Settings. In ChatGPT, turn off "Improve the model for everyone" under Settings, then Data controls.

Is a bigger context window always better?

Not always. A bigger context window lets you paste a whole contract or codebase at once, but models tend to miss details buried in the middle of long inputs. Every token you send also counts against your usage or your bill, so a focused excerpt often does better than the whole file.

Does ChatGPT show ads?

On the cheaper plans it can. OpenAI says ads may show on the Free plan in some countries and on Go, the budget plan between Free and Plus. Claude's plan table has nothing between Free and Pro.

Which is better value through the API, Claude or GPT models?

GPT-6.1 Sol, in the middle tier. It lists at the same per-token rates as Sonnet 5.5, so its much lower cost per index task means it uses far fewer tokens to finish one. Opus shows the same pattern at the top: cheaper per token than Astra, dearer per task.

Is it worth paying for both Claude and ChatGPT?

Only if each covers work the other can't. A common split is ChatGPT for images and Claude for long writing, which costs $40 a month. Both have free plans, so many people pay for one and keep the other free. The AI stack cost calculator adds any other subscriptions to that total.

Sources: Anthropic's and OpenAI's plan pages, help-center articles and model announcements, plus the Artificial Analysis and Arena leaderboards, all read on 20 October 2026. Where a benchmark comes from a vendor, the text says so.