claude alternatives: what to use instead for chat, coding and writing

the best Claude alternatives in September 2026 are ChatGPT for an all-round chat app, OpenAI Codex or Cursor instead of Claude Code, Gemini if you live in Google apps, and open models like gpt-oss or Qwen if you want something free that runs on your own machine. which one fits depends on what you used Claude for, so this guide is split by job. prices and free tiers were checked on the vendors' own pages on September 27, 2026.

why people look for a claude alternative

the usual reasons are limits, price and access. Claude's free plan has no Claude Code, Claude Design or Research mode: those start with Pro at $20 a month, and heavy users end up on Max at $100 or $200 (Claude pricing). usage limits are the most common complaint – we covered the mixed reaction to Opus 4.7's 50% more tokens and 5-hour limits. and when Anthropic goes down, everything built on it stops, as it did in the outage that hit claude.ai, the API and Claude Code at once.

you don't have to leave Claude completely. a lot of people keep it for one job and use something else for the rest – that's how we work too (see what we use).

the best claude alternatives for chat

chatgpt

our pick as the all-round replacement for the Claude app. plans: Free, Go at $8 a month ("this plan may include ads"), Plus at $20 and Pro from $100 (OpenAI pricing). free and Go users get GPT-5.6 Luna by default, paid plans get GPT-5.6 Sol with a thinking slider, and Pro, Business and Enterprise get GPT-6 Pro, powered by GPT-6 Astra (OpenAI help).

what Claude users will recognise: projects on every plan, memory of past chats, file uploads, deep research and apps that connect to your tools. the main thing Claude doesn't do is generate photos and illustrations – it draws charts and diagrams as code instead (Claude help); both apps have voice. limits on the free plan are real – uploads, images and voice are capped, and in the Instant model the context window is 27K tokens on Free against 128K on Pro.

gemini

the pick if your work lives in Google Docs, Gmail and Chrome. the free plan includes Gemini 3.6 Flash and "varying access" to Gemini 3.1 Pro, image generation, Deep Research, Gemini Live voice, Canvas and Gems. paid tiers: Google AI Plus $4.99 a month, AI Pro $19.99, AI Ultra from $99.99 (Google subscriptions). AI Pro adds Gemini in Gmail and Docs, video generation, 5 TB of storage and a 1M-token context window – a good fit for long documents, though some Claude models now offer up to 1M tokens too (Claude pricing).

perplexity

our pick to replace Claude's research mode: search answers come with inline citations. the free tier gives practically unlimited basic searches and a small number of Pro searches; choosing between advanced models from several labs, Claude among them, needs a paid plan (Perplexity help). Pro is $20 a month or $200 a year, Max $200 a month (Perplexity pricing, Perplexity Max).

mistral vibe (formerly le chat)

Le Chat was renamed Vibe on August 12, 2026 and now has Chat, Work and Code modes (Mistral help). the free plan includes limited messages and web searches, image generation, 100+ connectors and $10 a month in API credits; Pro is $14.99 a month, with a $5.99 offer for verified students new to Vibe, for up to 12 months (Mistral pricing). a good choice if you want a European provider and custom MCP connectors (beta) – the same kind of connectors we explain in how to build an MCP server.

grok, microsoft copilot, kimi, deepseek and qwen

• Grok – live search across the web and X, image and video generation. free to try; SuperGrok Lite is $10 a month and SuperGrok $30 (Grok plans).
• Microsoft Copilot – a limited free version; the paid Copilot now comes inside Microsoft 365, from Personal at $9.99 a month, with Copilot in Word, Excel and PowerPoint (Microsoft).
• Kimi – Moonshot's app on Kimi K3 with slides, docs, sheets and deep research. free tier; Plus is $15 a month billed annually (Kimi pricing). in our own agent test, Kimi K3 outperformed Claude Opus 5 driving the exact same agent.
• DeepSeek and Qwen Chat – free chat apps from Chinese labs; Qwen describes its app as "free to use, open to all" (Qwen). no paid consumer plan is listed on deepseek.com. both labs also publish open-weight models, such as Qwen3.8-27B, that you can run yourself (see free and open-source).

claude code alternatives

claude Code isn't included in Claude Free: subscription access starts with Pro, or you pay per use through the API (Claude pricing, Claude Code authentication). the alternatives split into three groups: other companies' coding agents, editors with agents built in, and open-source agents where you bring any model.

openai codex

the closest match to Claude Code: a terminal agent, IDE extensions, a desktop app and cloud tasks; Codex CLI is open source under Apache-2.0 (GitHub). Codex comes with ChatGPT plans: Free and Go get it in the desktop app as it rolls out, while the CLI, IDE extensions and web start with Plus; it supports MCP servers (Codex pricing, MCP docs). we run it every day next to Claude Code – details in what we actually use. we've also compared the underlying models in our own tests (set-ups differ between tests, see each article), for example Opus 5.5 vs Fable 5.1 vs GPT-6 Astra on three film worlds and GPT-6 Astra vs Fable 5.1 on a Goldberg machine.

cursor and github copilot

if you'd rather stay in an editor. Cursor has a free Hobby tier with limited agent requests and Pro at $20 a month, with models from OpenAI, Anthropic, Google and SpaceXAI (Cursor pricing, Cursor models). GitHub Copilot's free plan includes the CLI and agent mode; Pro is $10 a month, and Pro+ ($39) can hand tasks to third-party coding agents, including Claude and Codex, in preview (Copilot plans). both support MCP (Cursor MCP, Copilot MCP).

antigravity replaces free gemini cli access

if you planned to use Gemini CLI for free: for unpaid and Google One users it was replaced by Antigravity CLI on June 18, 2026; Gemini CLI still works with paid API keys (Google Developers Blog). Antigravity's individual plan costs nothing and includes Gemini Flash models, Gemini 3.1 Pro, Claude Sonnet and Opus 4.6 and gpt-oss-120b with weekly rate limits (Antigravity pricing).

devin desktop, formerly windsurf

Windsurf was renamed Devin Desktop on June 2, 2026; Cognition also offers Devin CLI and Devin Cloud. the free plan has a light agent quota, Pro is $20 a month and includes Devin Cloud, and it supports MCP servers (Devin Desktop FAQ, Devin pricing, Devin MCP).

open-source agents: opencode, cline, zed, aider

• OpenCode – MIT-licensed terminal agent (license) with 75+ model providers and local models through Ollama or LM Studio; OpenCode Go is an optional $10 subscription to open models, not a licence fee (OpenCode Go). it works with ChatGPT Plus/Pro and GitHub Copilot subscriptions; its docs note that Anthropic prohibits using a Claude Pro subscription inside it (OpenCode docs).
• Cline – the VS Code extension and CLI are free and open source (Apache-2.0); you bring your own key or a local model. the JetBrains plugin has its own licence, so check its terms first (Cline pricing, JetBrains licence, installing Cline).
• Zed – an open-source editor with an agent, free with your own API keys, Pro $10 a month (Zed pricing); it can also host other agents such as Claude Code through the Agent Client Protocol (Zed).
• Aider – still on many lists and still auto-commits every change to git, but its latest tagged GitHub release, v0.86.0, is from August 2025 (releases). if you need MCP, pick one of the three above.

or keep claude code and swap the model

several labs publish Anthropic-compatible endpoints, so Claude Code itself can run on their models. DeepSeek documents ANTHROPIC_BASE_URL=https://api.deepseek.com/anthropic (DeepSeek docs), Z.ai does the same for GLM-5.3 (Z.ai docs), and Ollama runs it on a local model with ollama launch claude, recommending at least 64K context (Ollama docs). Anthropic itself doesn't support routing Claude Code to non-Claude models (Claude Code docs), so treat this as a do-it-yourself setup. we haven't measured the savings inside Claude Code; for a sense of model prices, in a separate API test DeepSeek V4.1 Flash ran all four three.js scenes for $0.15.

claude alternatives for writing

claude is still the one to beat for creative writing on human votes. on the LMArena creative writing leaderboard (updated September 25, 2026) the top three are Claude models; the best non-Anthropic entry is Gemini 3.7 Flash (high), fourth and still marked preliminary, and the best GPT entry is GPT-5.6 Sol (xhigh) at fourteenth (LMArena). EQ-Bench's creative writing test, which is judged by a model rather than people, puts GPT-6 Astra first, Claude Fable 5.1 second and, among open models, Kimi K3 and GLM-5.3 at fifth and sixth (EQ-Bench, leaderboard data, checked September 27).

in practice: for fiction and voice, try Gemini first; for structured writing – briefs, emails, summaries – write a specific prompt before judging a model. in our small prompt test (Claude Opus 5.5 in Claude Code, GPT-6 Astra in Codex, one run per task) both scored full or near-full marks on a checklist written in advance once the prompt was good (how to write AI prompts); in that test the prompt mattered more than the model.

claude design and cowork alternatives

claude Design turns a chat into designs and prototypes (slides and docs now have their own Slides and Docs tools); it launched on April 17, 2026 for paid plans (Anthropic, Claude help). alternatives:

• v0 by Vercel – prompt to UI and app; free plan with $5 of monthly credits (v0 pricing).
• Lovable – prompt to a working web app; 5 free credits a day up to 30 a month, Pro at $25 a month (Lovable).
• Figma Make – inside Figma; the free Starter plan can try it with limits, full use needs a Professional Full seat at $16 a month billed annually or $20 monthly (Figma Make FAQ, Figma pricing).
• Google Stitch – a Google Labs "AI design canvas" that turns text into UI (Google Labs).

claude Cowork – the agent that works across your files and tools – is now part of the main Claude app ("Claude Cowork is now just Claude", Claude). the alternatives are other long-running agents:

• ChatGPT Work – replaced ChatGPT agent (OpenAI help); built for longer multi-step work, can use local files and apps in the desktop app with your permission, supports scheduled tasks, and shares a usage budget with Codex (OpenAI help).
• Gemini Spark – works in the background even when your devices are off; AI Pro and Ultra, personal accounts 18+ with Keep Activity on, not available in the EEA, UK, Switzerland and Nigeria (Google, Gemini help).
• Perplexity Comet – an AI browser with a free version (Comet); the Computer agent comes with Pro and Max (Perplexity help).
• Manus – a free plan and paid plans from $20 a month (Manus help).

free and open-source claude alternatives

the cheapest way off Claude is a free tier: ChatGPT Free, Gemini, Mistral Vibe, Qwen Chat, DeepSeek and Kimi all let you chat without paying, and GitHub Copilot has a free coding tier.

if you want your chats to stay on your machine, run an open model locally with cloud features and outside tools switched off. candidates:

the easiest way to run them is Ollama or LM Studio. the big open models – DeepSeek V4.1, GLM-5.3, Kimi K3 – are too large for a laptop but cheap through their APIs. if you're thinking about hardware for bigger ones, we compared a Mac Studio and an RTX 5090 for local AI.

how much usage you actually get for the money

this is the question every "claude alternatives" list skips, because vendors don't publish comparable numbers. claude, Codex and Gemini use rolling 5-hour and weekly limits described as multiples ("5x Pro") or estimated ranges, not guaranteed amounts of work. what follows is the official part where it exists, and then what users who measured their own usage report. the community numbers are single accounts' estimates, change with promotions and model choice, and aren't guarantees – read them as a rough sense of scale from recent reports, each dated below.

what vendors publish

Cursor and Copilot state the included usage in dollars; Claude, Codex and Gemini give multiples or ranges, so the real amount depends on the model and the task.

what users report

comparable official numbers aren't published, but recent reports from people who logged their own usage give roughly this picture:

• claude Max. one Max 20x user who sampled the usage meter for over three weeks in June and July, mostly on Opus in Claude Code, could use about six full 5-hour sessions a week before the weekly cap. another user who priced every token at API rates got the equivalent of about $790 a week on Max 5x in July and about $1,700 a week on Max 20x in August – during a promotion that raised Claude Code's weekly limits by 50%. by that measurement Max 20x gave about 4.4 times the 5-hour allowance of Max 5x but only about 2.2 times the weekly one; a separate September measurement put the weekly gap at 1.5 times. the promotion ran until September 13 and was replaced by a permanent 25% increase; the same user's early extrapolation after that – two days of data covering 29 percentage points of weekly usage, on Fable 5.1 with Max 5x – was about $470 a week.
• claude, time of day. one September test found identical requests draining the 5-hour meter about 1.4 times faster on weekdays between 12:00 and 18:00 UTC. it's a single user's finding, not a published rule.
• ChatGPT Plus in Codex. in September a Plus user hit the 5-hour limit on GPT-6 Astra after about 980,000 tokens, roughly $3.55 at API prices; another, on a Business plan with GPT-5.6 Sol at xhigh effort, found one full session used about 16% of the weekly allowance – around six sessions a week – and said Plus behaved similarly.
• ChatGPT Pro $100 in Codex. a user tracking spend against the meter in July estimated about $600–675 of API-priced usage a week.
• Gemini in Antigravity CLI. a September benchmark on Google AI Pro counted about six full 5-hour windows a week and estimated a capacity of roughly 95 multi-step coding tasks a week with Gemini 3.8 Flash or 180 with Gemini 3.1 Pro, from its own task suite. Ultra wasn't measured.
• Cursor. in September, one Pro+ user reported about 1.8 billion tokens in a month, roughly 80% on Grok 4.6 and 20% on Composer 2.5; another reported about 2.9 billion, mostly through Auto, without knowing which models Auto picked. that doesn't translate into Claude-class tokens: cache reads and cheaper models inflate the count.
• GitHub Copilot Pro+. a September user spent the whole $70 monthly credit allowance in about two weeks of coding work.

what this means for you. these individual estimates suggest that for some heavy coding workloads Claude Max and ChatGPT Pro are worth several times their price at API rates; they don't establish a typical return or a winner. on a $20 budget, Cursor and Copilot are the more predictable ones, since the included balance is stated in dollars; how far Claude Pro or ChatGPT Plus goes depends on the model and the task. the honest way to compare is to run your own typical task on each plan and watch the meter.

moving from claude: what comes with you

chats, projects and memory don't move by themselves. export what you need from Claude before you cancel, and check whether the new app can import it: Gemini, for example, documents importing memory and chats from other AI apps, with account and region limits (Gemini help). MCP servers you connected to Claude Code can usually be reused in Codex, Cursor, Copilot, OpenCode and Zed, since they all support MCP, but you set up each connection and its auth again and should check the transport (stdio or HTTP) each tool supports – our MCP server guide shows the connect commands for Claude Code, Codex and Gemini. prompts are the part worth rewriting rather than copying: each model reads instructions a little differently.

what we actually use

this part is our own experience, not a summary.

• Codex next to Claude Code, every day. we write code in Claude Code and send it to Codex (GPT-6 Astra through a ChatGPT subscription) for review and fact-checking. on September 27, 2026 its review of our own publishing scripts found a shell-injection bug that could have allowed commands to run on our server; its fact-check of our MCP guide caught a wrong claim about Gemini before it went live. Codex also generates our article covers, including this one.
• what our subscription gave us in Codex. on September 27, 2026, on our account, Codex CLI ran GPT-6 Astra, while requests for GPT-6 Sol and GPT-6 Luna returned an error saying they aren't supported with a ChatGPT account – even though OpenAI's pricing page lists them for Plus, so your account may differ. claude Code on a Claude subscription let us pick Opus 5.5, Fable 5.1 or Sonnet 5.
• same prompts, two models. for our prompt guide we ran identical prompts through Claude Opus 5.5 and GPT-6 Astra in clean setups. with well-written prompts Claude scored 18 of 18 and GPT-6 Astra 17 of 18; on the coding task both passed all 23 hidden tests. one run per task, so it's a small sample, and it says nothing about moving projects, memory or integrations between apps.

how we picked

we started from what people actually search for – "claude alternatives" for chat, coding, writing and design, and "free" and "open source" versions – and covered each job with the tools that appear on most top-ranking lists plus the ones those lists missed after 2026's renames (Vibe, Devin Desktop, Antigravity, ChatGPT Work). prices, free tiers and features come from the vendors' own pricing pages, help centers and docs, checked on September 27, 2026; a few pricing pages load prices dynamically and may show different figures by region. rankings come from the named leaderboards. "Best" and "our pick" are editorial choices based on that documentation, not head-to-head tests, and we only say "we" about things we ran ourselves.

faq

what is the best alternative to claude?

for most people, ChatGPT: it covers chat, files, projects, research, images and voice, and it has a free tier. for coding, OpenAI Codex is the closest thing to Claude Code; its CLI comes with ChatGPT Plus.

is there a free claude alternative?

yes. ChatGPT, Gemini, Mistral Vibe, Qwen Chat, DeepSeek and Kimi all have free chat; GitHub Copilot has a free coding tier with the CLI and agent mode, and Codex is rolling out in the desktop app for ChatGPT Free. for no subscription at all, run gpt-oss-20b or a small Gemma 4 locally with Ollama – you only need the hardware.

what is the best claude code alternative?

OpenAI Codex if you want a terminal agent like Claude Code, Cursor or GitHub Copilot if you want an editor, and OpenCode or Cline if you want open source with any model.

which ai is better than claude for writing?

on the LMArena creative writing board (September 25, 2026), Claude models hold the top three places and Gemini 3.7 Flash is the closest rival; on EQ-Bench's model-judged test GPT-6 Astra ranks first. try both on your own writing.

which subscription gives the most usage for the money?

vendors don't publish comparable numbers. Cursor and GitHub Copilot state the included usage in dollars; Claude, Codex and Gemini use 5-hour and weekly limits described as multiples or ranges. individual users' estimates suggest heavy coding on Claude Max or ChatGPT Pro can be worth several times the price at API rates, but results vary a lot by model and promotion.

can i use claude code with other models?

DeepSeek, Z.ai (GLM) and Ollama document Anthropic-compatible endpoints for it, but Anthropic doesn't support routing Claude Code to non-Claude models, so it's at your own risk.

what happened to gemini cli and windsurf?

for unpaid users Gemini CLI was replaced by Antigravity CLI on June 18, 2026. Windsurf was renamed Devin Desktop on June 2, 2026.

sources

• claude pricing, Max plan, Fable on your plan, Claude Code: LLM gateway, Claude Design launch, Claude Cowork, Claude Code authentication, Pro tiers, Design, Slides and Docs, can Claude produce images
• ChatGPT pricing, GPT-5.6 and GPT-6 Pro in ChatGPT, ChatGPT Work and Codex, ChatGPT agent, Codex pricing, Codex MCP, Codex on GitHub
• Google AI subscriptions, Gemini usage limits, Gemini Spark, Gemini Spark help, Gemini import, Gemini CLI to Antigravity CLI, Antigravity pricing, Google Labs
• Perplexity pricing, Perplexity plans, Perplexity Max, Comet
• Le Chat is now Vibe, Mistral pricing
• Grok plans, Microsoft Copilot for individuals, Kimi pricing, Qwen
• Cursor pricing, Cursor included usage, Cursor models, Cursor MCP, Copilot MCP, Devin Desktop FAQ, Devin pricing, OpenCode license, OpenCode Go, Zed pricing, GitHub Copilot plans, OpenCode providers, Cline pricing, installing Cline, Zed external agents, Aider releases
• DeepSeek coding agents, Z.ai for Claude Code, Ollama for Claude Code
• LMArena creative writing, EQ-Bench creative writing (data)
• v0 pricing, Lovable plans, Figma pricing, Figma Make FAQ, Manus pricing
• gpt-oss-20b, Gemma 4, Muse Glimmer 30B, Qwen3.8-27B, Ollama, LM Studio