ChatGPT Alternatives by Job: Writing, Research, Code

I still pay $20 a month for ChatGPT Plus, and I only open something else when one job has outgrown it. For writing and long documents that job goes to Claude at the same price, and each tool below takes over one other job where ChatGPT is weaker, pricier or less private.

Writing and long documents: Claude

Claude Pro is $20/mo, the same as ChatGPT Plus, so the reason to switch is output quality. Claude Opus 4.8 writes cleaner prose than GPT-5.6 out of the box and keeps long documents coherent past the point where ChatGPT starts hedging. Its coding agent is now a product in its own right.

What you give up: native image generation, anything like the Custom GPTs marketplace, and predictability, because the rolling 5-hour usage window can cut off a heavy session with no warning. Team and Enterprise plans are opted out of training by default; individual plans are not. The verified Claude profile has the rest.

Workspace work, or no budget at all: Gemini

Google AI Plus starts at $4.99/mo and AI Pro is $19.99/mo (checked 2026-08-25, gemini.google/subscriptions). The free tier includes image generation and Deep Research and needs no credit card. If your work already sits in Gmail, Docs or Drive, the integration earns its place.

Gemini 3.1 Pro is close to GPT-5.6 for most tasks and clearly ahead on long-context summarization. Privacy is the weak spot: it trains on your chats unless you turn off Keep Activity in your Google Account settings, and the default retention window is 18 months.

Cited research: Perplexity

If your complaint about ChatGPT is that the citations aren’t real, this is the switch. Perplexity Pro is $17/mo billed annually or $20/mo monthly (checked 2026-08-25, perplexity.ai/pro). Pro searches route across its in-house Sonar model plus Claude, Gemini, GPT-5.6 and Kimi, so you get four frontier models in a search-first interface with a citation on every claim.

The free tier is thin: three Pro searches a day and one Research query a month. If citations don’t matter to you, ChatGPT Plus is the better $20.

Coding and reasoning on a budget: DeepSeek

DeepSeek’s savings show up on the API. DeepSeek-V4-Flash costs $0.14 per million input tokens (checked 2026-08-25, platform.deepseek.com/pricing), against $2.50-$15 per million on OpenAI’s flagship tiers. The consumer app at chat.deepseek.com costs nothing and advertises no message cap, and the open-weight versions can be self-hosted.

The catch doesn’t bend. DeepSeek’s own privacy policy says data is processed and stored in the People’s Republic of China, prompts train the model by default, and there’s no image generation. Keep sensitive data out of it.

Work inside Word, Excel and Outlook: Microsoft Copilot

Copilot only earns a switch if you already pay for Microsoft 365. It’s free with a Microsoft account, and heavier use comes bundled in M365 Personal ($9.99/mo), Family ($12.99/mo) and Premium ($19.99/mo) instead of a standalone consumer plan. Inside Word, Excel, PowerPoint, Outlook and Teams it does useful work.

Outside those apps the gaps show. Microsoft doesn’t name the underlying model, publishes no free-tier message caps, offers no clean data export, and its consumer privacy terms allow training “in some markets” unless you opt out. Country availability is just as fuzzy.

Private or offline work: Ollama

Ollama is for work that shouldn’t leave your computer. It runs open-weight models such as Llama 3.3, gpt-oss, DeepSeek-V4 and Mistral on your own hardware, free and without limits, and sends your prompts nowhere. You can also point Claude Code or Codex at those local models and skip the API bill.

The cost is hardware: the useful models need a recent Apple Silicon Mac or a GPU with 24GB+ VRAM. There’s no image generation, and the output is only as good as the open model you load.

Price, free tier and training defaults compared

Tool	Best for	Entry paid	Free tier does real work	Trains on your data by default
ChatGPT	The default	$20/mo	Yes, limited	Yes (opt-out available)
Claude	Writing, long context	$20/mo	Yes	Yes on individual (Team/Ent no)
Gemini	Workspace, free tier	$4.99/mo	Yes, generously	Yes (Keep Activity toggle)
Perplexity	Cited research	$17/mo	Barely	No on Enterprise
DeepSeek	Cheapest coding	Free consumer / API	Yes, unlimited chat	Yes, data hosted in PRC
Copilot	M365 integration	Bundled in M365	Yes	Yes “in some markets”
Ollama	Local privacy	Free (your hardware)	Yes, unlimited	Never

All prices and policies checked 2026-08-25 on each vendor’s own pages.

Tools that didn’t win a job

• Grok costs more than Plus (SuperGrok is $30/mo, SuperGrok Heavy $300/mo), and xAI publishes no clean quota or model-version information. The X-search hook is real but narrow; the field is in Grok alternatives.
• Meta AI is free until Meta decides otherwise, and paid tiers are already testing in three countries. It trains by default with no clean opt-out and no incognito mode, which makes it fine for casual chat inside WhatsApp and little else.
• Mistral Vibe is strong on EU data residency, but the rebrand from Le Chat is still settling and its integration ecosystem is smaller. I treat it as a specific-use pick, as the Mistral alternatives page does.
• Lumo, Proton’s assistant, has zero-access encryption but no coding agent, no voice mode and no named model. It belongs on a privacy shortlist, not in daily use.
• Kimi has long context and cheap coding, but Moonshot AI offers no training opt-out at all. If training defaults are why you’re leaving ChatGPT, it takes you backwards.
• Poe bundles many models under one subscription, but each bot carries its own privacy terms, and for real usage the bill comes close to ChatGPT plus one competitor paid directly.
• HIX.ai meters credits from the first message, Andi has no image or file upload, and Indus is India-only behind a waitlist.

Keep ChatGPT and add one tool

Replace ChatGPT outright only when your main job is cited research, private local work or Microsoft 365. For writing, run Claude next to it: same price, different strengths. DeepSeek replaces it on the API only if China-hosted data is acceptable to you.

If ChatGPT still handles 80% of your tasks, keep it and add the one tool that covers the gap. That’s my setup.