Claude alternatives for coding: costs, catches and free options

If you want something cheaper than Claude Code, or just different, start free. Kiro’s free tier still gives you a Claude model. GitHub Copilot Free covers everyday autocomplete. Devin Desktop is what Windsurf became. Pay around $20 a month only when a free quota is clearly slowing down paid client work. Cursor is still a strong editor, but on its new plans Claude models are billed at API rates on top of the subscription, so read the fine print before you switch.

What AI speeds up here

The price table, checked October 2026

Coding tool pricing has moved a lot this year. Windsurf was renamed, Cursor restructured its plans, and Gemini CLI stopped serving most individual users. The sections after the table explain the catches.

The Claude figures come from Anthropic’s pricing page, which also confirms that Claude Code comes with every paid plan but not with Free. The ChatGPT tiers come from news reporting, not OpenAI’s own pages. For the wider freelancer toolkit beyond coding, see our roundup of AI tools for freelancers, with free tiers and costs.

Kiro is the cheapest confirmed way to keep a Claude model

For a side hustler, Kiro has the most interesting free tier on this list. Kiro’s pricing page says the free plan is permanent and includes Claude Sonnet 4.5, plus open-weight models such as Qwen3 Coder Next, DeepSeek 3.2 and MiniMax M2.1, all with rate limits. You sign up with a social login or an AWS Builder ID.

Paid plans can buy extra credits at $0.04 each, in packs from $5 (125 credits) up to $100. Those add-on credits expire 12 months after purchase. If you signed up with a social login or Builder ID, you get $20 off your first upgrade. Watch how quickly your own tasks use up the 50 free credits, and treat the free tier as a test drive, not an allowance you can plan a deadline around.

Cursor changed its plans this month

Cursor now sells its plans as “The X Subscription,” which bundles Grok, Cursor and X. According to Cursor’s models and pricing docs, paid plans share one weekly usage pool across Grok and Cursor. Claude, GPT and Gemini sit outside that pool:

1. On Free and Plus, you can’t use third-party models at all.
2. On Premium and above, every Claude request is charged at the model’s API list price from the first one, paid from prepaid credit.
3. Teams and Enterprise also pay a “Cursor Token Rate” of $0.25 per million tokens on top.

Older plans (Pro, Pro+, Ultra, Start and Teams Standard/Premium) are now legacy plans that “keep working as they are,” so existing subscribers aren’t forced to move. The change hits new sign-ups. If your main reason for picking Cursor is Claude inside a polished editor, a new plan means paying twice: once for the subscription, then again for the tokens. Cursor makes the most sense now if you’re happy working mostly with Grok models.

Windsurf is now Devin Desktop

Cognition agreed to buy Windsurf in July 2025 and announced the rename to Devin Desktop on June 2, 2026. The old pricing URL now redirects to Devin’s pricing page. The free plan gives you unlimited Tab completions and inline edits, plus a light agent quota. Pro is $20 a month. Allowances refresh daily and weekly, and paid users can buy more at API pricing. Unlimited autocomplete on a free plan is useful for small fixes. Just see how fast you burn through the agent quota before you rely on it for a deadline.

Google moved its CLI users to Antigravity

Older guides often name Gemini CLI as the free choice, and that advice is out of date. Google’s developer blog says that on June 18, 2026, Gemini CLI and the Gemini Code Assist IDE extensions stopped serving requests for Google AI Pro and Ultra subscribers and for free “Gemini Code Assist for individuals” users. Those users were moved to Antigravity CLI. Gemini CLI still works with paid Gemini API keys, and enterprise Code Assist licenses are unchanged. The post gives no free quota or price for Antigravity CLI. If you already pay $19.99 a month for Google AI Pro for its 5 TB of storage, you get the Jules coding agent with it at no extra cost. That’s the strongest case for trying Google. Our ChatGPT, Claude and Gemini comparison covers the non-coding side of each plan.

Open-source tools: free to install, not free to run

Cline and Aider are both open source under Apache 2.0, and neither is tied to one AI company. Cline works with Anthropic, OpenAI or Google models, or with local models through Ollama or LM Studio. Aider “can connect to almost any LLM, including local models,” using your own API key from the provider. You can bring a Claude key, but then you pay Anthropic’s API rates on every request, with no monthly cap. Running a model on your own computer avoids that bill, but you’ll need fairly powerful hardware to run a model good enough for client work.

Pick a tool in five steps

1. Write down the coding work you do each week. Autocomplete while you type is cheap or free almost everywhere. Handing whole features to an agent is where quotas run out.
2. Start with one free tier that matches: Copilot Free or Devin Desktop Free for autocomplete, Kiro Free if you want to keep a Claude model.
3. Run the same task in two tools. Use the prompt below on a small project you own, then compare the plans, the diffs and how much quota each one used.
4. Upgrade only when a limit costs you billable time. Copilot Pro at $10 is the cheapest paid step here, unless you’re fine with Cursor Plus at $8 and Grok models only. Kiro Pro, Devin Pro and Claude Pro all cost $20.
5. Check the pricing page before each renewal, and avoid annual billing on a tool that has changed its plans recently.

Prompt you can copy

This works in any agent-style tool. It keeps the agent on one task and makes it show a plan before it edits anything, which also makes side-by-side comparisons fair.

You're working in an existing [LANGUAGE / FRAMEWORK] project for a client.

Task: [ONE SENTENCE DESCRIBING THE FEATURE OR BUG]
Files you may change: [LIST OF FILES OR FOLDERS]
Files you must not touch: [e.g. .env, database migrations, payment code]

Before editing anything:
1. Explain how the relevant code works today in 5 bullets or fewer.
2. List your assumptions and any questions you'd ask the client.
3. Propose the smallest set of changes that completes the task.
Then stop and wait for approval.

Once the plan is approved and you've made the edits:
- Summarize the diff file by file.
- Add or update tests for [EXPECTED BEHAVIOR].
- List anything you couldn't verify (library versions, API behavior, env variables).

Where AI falls over

• Every tool here sometimes calls functions that don’t exist or suggests outdated library versions. Run the code instead of just reading it.
• Agents “tidy up” files you never mentioned. The “must not touch” line helps, and reviewing the diff before you commit catches the rest.
• A test that checks the code against itself will always pass, so read the assertions.
• Daily and weekly allowances (Devin) and credits that don’t roll over (Kiro) can stop an agent halfway through a refactor. Commit often so you can pick up cleanly in another tool.
• Your favorite tool can change its plans while you’re in the middle of a client job. Keep your workflow portable: plain git, prompts saved in your own files, and no habits that only work in one editor.

The human-only part

What a client is really buying is a working result they can trust. Working out what they actually need, pushing back on scope creep, checking security and owning the fix when production breaks all fall to you. Those are the parts that justify your rate. Boilerplate pays least because AI makes it cheap. Always review and test AI-written code before you deliver it. If you’re still choosing what to offer, our AI business ideas from home guide splits each idea into what the AI does and what you do.

Two practical points. First, read your client contract before you paste their code into any AI tool, and never paste secrets or customer data. Second, if this is 1099 income, a subscription you need for the work can be a deductible business expense. IRS Publication 334 says the expense must be “ordinary” and “necessary,” and the personal share of mixed use is generally not deductible. Self-employment tax is 15.3%, and you file Schedule SE once your net self-employment earnings reach $400. This is general information, not tax advice, so check your own situation with a tax professional.

Try it in 20 minutes

Pick a small repo you own, not client code. Sign up for Kiro’s free tier and turn on GitHub Copilot Free in VS Code. Paste the prompt above into each one with the same small task, such as “add input validation to the signup form.” Approve both plans, then compare three things: which plan was more accurate, which diff stayed inside the files you allowed, and how much free quota each one used. By the end you’ll know which free tool fits the way you work, and whether you need a $20 plan at all.