On Super Bowl Sunday, February 8, 2026, Anthropic ran a 30-second spot. Not for Claude itself. For the idea that your AI shouldn't show you ads. "There is a time and place for ads," the broadcast version said. "Your conversations with AI should not be one of them."
The next morning, on February 9, ChatGPT started showing ads to free and Go-tier users.
Sam Altman called Anthropic's campaign "so clearly dishonest." The internet called it prescient. Either way, that one week captures the ChatGPT vs Claude divide better than any benchmark table could. These two companies have different philosophies about what AI should be, and those philosophies show up in their products.
I use both every day. Claude vs ChatGPT isn't a question of which is better. It's a question of what you're doing right now, what you're willing to pay, and what trade-offs you can live with. Here's where each one actually wins.
The 30-Second Answer (September 2026)
Use Claude if you write for a living, code for a living, or need to think through something complicated. Claude Opus 5.5 (released September 22, 2026) reads up to 1 million tokens, outputs up to 128,000 tokens in a single response, and produces text that sounds like a person wrote it.
Use ChatGPT if you need images, voice mode, unlimited free text chats, or the widest feature set in one app. GPT-5.6 Sol powers chat on paid plans, GPT-6 Astra (unveiled September 3) runs the heavy work, and OpenAI keeps shipping features faster than anyone can track. Image generation, the ChatGPT Work agent, scheduled tasks, an $8/month Go tier.
Use both if you're serious. The people getting the best results in 2026 aren't loyal to either platform. They switch based on the task.
Now the details.
Writing: Claude Wins, and the Gap Got Wider
ChatGPT vs Claude for writing is the comparison with the clearest answer. Claude's output sounds human. ChatGPT's output sounds like AI wrote it and then tried to hide the evidence.
OpenAI retuned GPT-5.6 Sol in August for more direct answers and tighter formatting, but the habits persist. "In today's ever-evolving landscape." "Let's dive in." "Here's the thing." You'll spend ten minutes scrubbing these phrases from every draft. Claude's defaults need less cleanup because the voice is more natural to begin with.
Where the gap widened: Claude lets you save a writing voice and reuse it. Anthropic is moving its custom styles into Skills, so a "casual blog voice" or "formal legal brief" preset becomes a skill you call up when a draft needs it. ChatGPT has custom instructions, but they're blunter. The difference is noticeable.
Claude also handles nuance better. Ask for "professional but warm" and you get professional but warm. Ask ChatGPT for the same thing and you get corporate with an exclamation mark.
Where ChatGPT catches up: bulk content. If you need 50 product descriptions or 100 social media captions, ChatGPT's speed and consistency across large batches still works. Writing blocks, which replaced Canvas on current models in May, let you edit a draft inline, which is useful for iterative work. And since August, ChatGPT Free gives you unlimited text chats on GPT-5.6 Luna, which makes it the cheapest option for casual writers who just need a first draft.
Verdict: Claude for anything that needs to sound human on the first draft. ChatGPT for high-volume content where you'll edit anyway.
Coding: The Agent War Changes Everything
A year ago, this section would've compared benchmark scores and called it a day. Not anymore. Something shifted in December 2025, and it changed how developers actually work.
Andrej Karpathy, former Director of AI at Tesla and a founding member of OpenAI, described it bluntly: a "phase shift in software engineering." He said his workflow flipped from 80% manual coding and 20% AI agents to 80% agent coding and 20% edits, in a matter of weeks. He called Claude Code "the first convincing demonstration of what an LLM Agent looks like" and the "biggest change to my basic coding workflow in ~2 decades of programming."
That's not a small claim from someone who built the AI team at Tesla.
Claude Code vs. Codex: Two Different Bets
Claude Code (Anthropic's CLI tool) connects directly to your codebase. It reads your files, writes code, runs your tests, creates commits, and pushes to GitHub. Boris Cherny, who built it at Anthropic, says 100% of his own code is now AI-written. Across the company, Anthropic puts the figure at 70-90%. It accounts for roughly 4% of all public GitHub commits, according to SemiAnalysis (February 5, 2026), and Anthropic said on February 12 that its run-rate revenue had grown past $2.5 billion. Accenture is training 30,000 professionals on it. Anthropic's docs put the average cost at about $13 per developer per active day across enterprise deployments.
OpenAI's answer is Codex, now running GPT-6 Sol and GPT-6 Astra on paid plans. Codex takes a different approach. Instead of a CLI in your terminal, it runs in a macOS desktop app, a CLI written in Rust (open source), a web interface inside ChatGPT, IDE extensions for VS Code, Cursor, and JetBrains, and on iOS. It's also on ChatGPT's free plan for quick coding tasks, and it shares one usage allowance with ChatGPT Work, OpenAI's agent. Claude Code needs a paid plan.
Here's the honest difference. Claude Code is more reliable right now. Developers report that it works on the first try more often, uses 5.5x fewer tokens than Cursor for the same task, and integrates into existing workflows without friction. Jaana Dogan, a principal engineer on Google's Gemini API team, posted that Claude Code built in one hour what her team had spent a year on (a distributed agent orchestrator). Jim VandeHei, the CEO of Axios and self-described "tech dope who knows nothing about coding," built four apps in eight hours on his phone using Claude.
Codex is more ambitious about the work around the code. GPT-6 Astra, the model behind it, is built for computer use and long multi-step workflows.
On raw benchmarks, they're nearly tied. Anthropic says Opus 5.5 performs at the level of Fable 5.1 on most work and leads in agentic coding. OpenAI calls GPT-6 Astra state-of-the-art on software engineering. The gap is smaller than the noise.
What separates them for heavy users is the limit. Anthropic says Pro and Max usage limits are "shared across Claude and Claude Code," so a long coding day can lock you out of chat until the window resets. On ChatGPT, Codex draws from the allowance it shares with Work.
What Karpathy Actually Said
His advice for working with AI coding agents is worth reading in full. The models act like "a slightly sloppy, hasty junior dev." They "really like to overcomplicate code." You need to "watch them like a hawk" for anything important. But he also says programming feels "more fun because a lot of the fill-in-the-blanks drudgery is removed."
His predictions: a "slopacolypse" of AI-generated code in 2026, plus a growing split between engineers who mostly liked coding and engineers who mostly liked building. Whether that excites or terrifies you probably determines which tool you'll pick.
Verdict: Claude Code for daily coding, reliability, and workflow integration. Codex for automation around the code, the free plan, and the OpenAI ecosystem. If you build software for a living, you likely want both, plus Cursor or Copilot on top.
Agents: ChatGPT Work vs Claude Cowork
Both companies now put an agent inside the main app. So if you're asking whether Claude has an agent mode like ChatGPT, the answer is yes, and as of this month it doesn't have a separate name.
Claude Cowork launched on January 12, 2026, as a research preview for people who don't code. On September 16, Anthropic announced that Cowork and chat are now one Claude. From any conversation, Claude can write a document with you in Claude Docs, build a deck in Claude Slides, or make visual work in Claude Design, and you can download the result as PowerPoint or PDF. By default it asks before it takes an action. You can loosen that so it keeps working and checks in only when it needs you. The merged app is rolling out to Pro and Max first, with Team and Free after.
ChatGPT Work is OpenAI's version. It's a separate tab next to Chat, built for longer multi-step jobs that end in a finished file. Scheduled Tasks let it run a job once, on a repeat, or when something changes. Plus subscribers get GPT-6 Astra inside Work, and Work draws from the same allowance as Codex.
| ChatGPT Work | Claude (formerly Cowork) | |
|---|---|---|
| Where it lives | Its own tab next to Chat | Inside every Claude conversation |
| Outputs | Docs, sheets, slides, web apps | Docs, slides, Design, Artifacts |
| Recurring jobs | Scheduled Tasks | Scheduled tasks, like a report every Monday |
| Default control | Plans the steps, pauses when it needs approval | Asks before acting, adjustable |
| Usage | Shared with Codex | Shared with chat and Claude Code |
If you want the mechanics behind both, our guide to what an AI agent is walks through how they plan, act, and check their work.
Verdict: ChatGPT Work for one place to run docs, sheets, apps, and Codex. Claude for agent output a client will read closely, though its merged app reaches Team and Free last.
Image and Video: ChatGPT Wins by Default
Claude can't generate images. ChatGPT can, and Images 2.5, rolling out to every tier including Free, produces genuinely impressive results. Product mockups, social media graphics, diagrams, memes. All without leaving the chat.
Video is gone from both. OpenAI shut down the Sora app and website on April 26, 2026, and the Sora API on September 24. Claude never had video. Our free AI tools guide covers the video generators people use instead.
Claude's counter: it writes better image prompts. If you use Midjourney or Stable Diffusion, Claude's descriptions are more detailed and produce better output. That's a workaround, not a feature.
Verdict: ChatGPT. No contest on images, and neither one makes video now.
Research and Analysis: Different Strengths
Both tools search the web. Both process uploaded documents. The difference is in what they do with the results.
Claude is better at synthesis. Upload five research papers and ask for a comparison. Claude will find contradictions, flag methodology differences, and build a structured argument. ChatGPT tends to summarize each paper in order without connecting the dots.
ChatGPT is better at breadth. Its web search is faster, pulls from more sources, and handles live data (stock prices, breaking news, sports scores) more consistently.
Claude's thinking is worth calling out. On Opus 5.5 and Fable 5.1 it's always on, and the model decides how long to reason based on how hard the question is. For complex analysis, that extra reasoning changes the quality of the output. It also reads more of your material at once. Opus 5.5, Fable 5.1, and Sonnet 5.5 each take about 555,000 words in one context.
Verdict: Claude for deep analysis of specific documents. ChatGPT for broad web research and current events.
The Trust Question: Ads, Yes-Men, and Who's Watching
This didn't used to matter for a feature comparison. Now it does.
ChatGPT ads launched February 9, 2026, initially on free and Go-tier users. By September they ran in more than 60 countries, still only on Free and Go. OpenAI framed it as a way to fund free access. Critics called it the beginning of the end. Zoe Hitzig, an OpenAI research scientist, resigned on February 9, citing the ad rollout. The worry isn't the ads themselves. It's whether ad revenue eventually shapes which answers ChatGPT gives you.
Then there's the yes-man problem. In April 2025, a GPT-4o update caused ChatGPT to agree with almost everything users said. It called a "shit on a stick" business idea genius. In one widely shared case, a user said they'd stopped taking psychiatric medication, and ChatGPT responded "I'm proud of you for speaking your truth." OpenAI rolled the update back within days, but the incident stuck.
Claude isn't immune to issues. It's sometimes too cautious, refusing requests that ChatGPT handles without problems. But Anthropic's positioning is clear: no ads, no third-party GPTs that might introduce misaligned incentives. Whether you find that reassuring or limiting depends on what you value.
The bottom line: if you're using AI for work where accuracy matters (legal, medical, financial), the trust architecture behind the model matters as much as the model itself.
Pricing: A New Tier Changed the Math
| Plan | ChatGPT | Claude |
|---|---|---|
| Free | Yes (unlimited text, with ads) | Yes (limited, no ads) |
| Go | $8/month in the US (with ads) | N/A |
| Plus / Pro | $20/month | $20/month, or $17/month billed yearly |
| Team | Business: $25/user/month, $20 billed yearly | Team: $25/seat/month, $20 billed yearly |
| Top tier | $100 or $200/month (Pro) | $100 or $200/month (Max 5x, Max 20x) |
ChatGPT's Go tier ($8/month in the US) gives you GPT-5.6 Luna with ads. That's the cheapest paid tier from either company. If you're a casual user who just wants better-than-free answers, Go is hard to beat on price.
At the $20 level, they match exactly. Claude drops to $17/month if you pay yearly. Neither plan gets you the top model without limits. Plus gets GPT-6 Astra only inside Work and Codex, not in regular chat, and Claude Pro reaches Fable 5.1 only through paid usage credits.
The top is where the math changed. Both companies now sell $100 and $200 tiers. ChatGPT Pro $100 gives you 5x the Codex usage of Plus and Pro $200 gives 20x, but OpenAI has paused new sign-ups for the $200 tier since September 10. Claude Max comes as Max 5x ($100) and Max 20x ($200). Heavy users hit rate limits on both platforms. The difference is what those limits cover, since Claude chat and Claude Code draw from one pool.
On the API, Opus 5.5 costs $4 per million input tokens and $20 per million output tokens.
Verdict: ChatGPT cheapest at the bottom (unlimited free text, Go at $8). Same at $20, with Claude cheaper billed yearly. Same prices at the top, where usage rules decide.
Features Gap (September 2026)
ChatGPT has:
- Image generation (Images 2.5, on every tier)
- Voice mode (GPT-Live-1 on paid plans, GPT-Live-1 mini on Free)
- Unlimited text chats on the free plan
- ChatGPT Work (agent for docs, sheets, slides, and web apps)
- Scheduled tasks ("Every Friday, summarize my week")
- $8/month Go tier
- Codex (coding agent, included on Free for quick tasks)
Custom GPTs are on their way out. OpenAI is retiring them in favor of plugins, personal plans can no longer create new ones, and existing GPTs are scheduled to stop running on December 11, 2026.
Claude has:
- Artifacts (live code and content previews in chat)
- Skills (saved writing voices and workflows, which replace custom styles)
- Claude Code (CLI agent for software development)
- 1M-token context window on Opus 5.5, Fable 5.1, and Sonnet 5.5
- Adaptive thinking (always on in Opus 5.5 and Fable 5.1)
- Claude Cowork (AI file and task agent for non-developers, launched January 12, 2026, merged into the main Claude app on September 16)
- Voice mode, including on the free plan
- 128,000-token output (Opus 5.5)
- No ads on any plan
That context window matters more than most people realize. Claude's 1M-token models handle roughly 555,000 words, per Anthropic's docs. OpenAI doesn't publish one context figure for ChatGPT's chat models, so test your longest document in both before you commit. If you work with large codebases, legal contracts, or research papers, Claude's large context means fewer "sorry, that's too long" errors.
The Dark Horse: Gemini
No ChatGPT vs Claude comparison in 2026 is complete without mentioning the third option. Google's free Gemini plan gives you Gemini 3.6 Flash, varying access to 3.1 Pro, and image generation and editing, which is the one thing free Claude can't do at all.
Gemini is strongest at multimodal tasks (mixing text, images, audio, and video in one conversation) and at tasks where Google's search infrastructure gives it a data advantage. It's weakest at the kind of extended writing and coding where Claude excels.
If you're already in the Google ecosystem (Workspace, Android, Chrome), Gemini integrates more naturally than either ChatGPT or Claude. Worth trying, especially on the free tier. Our Gemini vs ChatGPT breakdown goes through it task by task.
When to Use ChatGPT
- You need images generated
- You want voice conversations on your phone
- You're doing broad web research on current topics
- You want unlimited text chats without paying
- You want the cheapest paid option ($8/month Go tier)
- You want scheduled tasks next to Codex and image generation (ChatGPT Work)
- You want a coding agent on the free plan (Codex)
When to Use Claude
- You're writing anything that needs to sound human
- You're coding and want reliable, first-try results
- You're analyzing long documents (contracts, codebases, research papers)
- You want the model to reason through hard analysis by default (adaptive thinking)
- You care about the ad-free model
- You need CLI-level codebase integration (Claude Code)
- You want an agent that asks before it acts (Claude, formerly Cowork)
- You're working with documents of several hundred thousand words (up to about 555,000)
The Honest Answer: Use Both
The "best AI chatbot" question assumes you have to pick one. You don't. $40/month gets you both standard plans. $28/month gets you ChatGPT Go plus Claude Pro. $0 gets you both free tiers, which are enough for casual use.
Here's what that looks like in practice. I write first drafts in Claude because the output needs less editing. I generate images in ChatGPT because Claude can't. I do deep code work with Claude Code in my terminal. I use ChatGPT for quick web lookups and voice mode in the car. I send long documents to Claude because the context window handles them. I use Codex when I need an agent to run for hours on a refactoring job.
The best ChatGPT alternative isn't Claude. The best Claude alternative isn't ChatGPT. The actual answer is to use both for different things.
And if you want prompts that work across both platforms, browse the prompt library. Every prompt there works with ChatGPT, Claude, and Gemini. Good prompts are model-agnostic. The techniques transfer.
Frequently Asked Questions
Is Claude better than ChatGPT? For writing and coding, yes. Claude produces more natural text and more reliable code. For image generation, voice mode, free usage, and feature breadth, ChatGPT wins. For research, it depends on whether you need depth (Claude) or breadth (ChatGPT). Neither is universally better.
Should I switch from ChatGPT to Claude? Don't switch. Add Claude for writing and coding work. Keep ChatGPT for images, voice, and web browsing. The free tiers overlap at zero cost. Both standard plans together are $40/month.
Is ChatGPT or Claude better for coding in 2026? Claude Code is more reliable for daily development work. Karpathy called it the "first convincing LLM agent" and said it changed his workflow more than anything in twenty years. OpenAI's Codex now runs GPT-6 Sol and Astra, comes with the free plan for quick tasks, and shares its allowance with ChatGPT Work. Claude Code shares its limit with Claude chat. Most professional developers use both, plus Cursor or GitHub Copilot.
Does ChatGPT have ads now? Yes. Since February 9, 2026, ChatGPT shows ads to free-tier and Go-tier ($8/month) users, and ads now run in more than 60 countries. Plus ($20/month), Business, Enterprise, and Pro ($100 or $200/month) subscribers don't see ads. Claude has no ads on any tier.
Is Claude or ChatGPT more expensive? Same price at $20/month, and Claude drops to $17/month billed yearly. ChatGPT is cheaper at the bottom (unlimited free text and an $8/month Go tier vs. no equivalent from Claude). At the top, both charge $100 and $200 a month, so the question becomes whose usage limits fit your work.
What about Gemini? Google's free Gemini plan runs 3.6 Flash with varying access to 3.1 Pro, plus image generation. It's strongest at multimodal tasks and anything that benefits from Google's search data. Weakest at extended writing and coding compared to Claude. Worth trying if you're in the Google ecosystem.
What are the disadvantages of Claude? No image or video generation. No $8/month budget tier. No free-plan coding agent (Claude Code needs a paid plan). Stricter rate limits on free and standard tiers. On paid plans, chat and Claude Code share one limit. Occasionally refuses requests due to more conservative safety filters.
What are the disadvantages of ChatGPT? Writing still sounds more "AI-generated" than Claude's. Code reliability trails Claude Code in daily use. Ads on free and Go tiers. The April 2025 yes-man incident raised trust questions. The top Pro tiers cost $100 and $200 a month, and new $200 sign-ups are paused. Feature overload can be overwhelming if you just want a good text conversation.
Does Claude have an agent mode like ChatGPT? Yes. Claude Cowork merged into the main Claude app on September 16, 2026, so any conversation can turn into a task that writes a document, builds slides, or makes a design. It's rolling out to Pro and Max first, then Team and Free. ChatGPT's equivalent is ChatGPT Work, a separate tab next to Chat.
What can Claude do that ChatGPT can't? Keep ads off every plan, including free, and write prose that needs less cleanup. Its 1M-token models also take about 555,000 words in one context, and OpenAI publishes no comparable figure for ChatGPT's chat models. ChatGPT's list runs the other way: images, unlimited free text, and a coding agent on the free plan.
Why do I lose context when I switch between ChatGPT and Claude? Because their memories are separate. Nothing you told ChatGPT carries over to Claude, or back. Keep a one-page project brief (goal, audience, decisions made, open questions) and paste it at the start of each new session. Before you switch mid-project, ask the current model for a recap using a summary generator prompt, then paste that recap into the other one.