Cursor vs Claude Code vs OpenCode: the UK indie hacker three-way for 2026

Yesterday we did the visual-builder three-way (Lovable vs Bolt vs Replit). Today is the IDE/agent layer underneath: Cursor, Claude Code, and OpenCode. The three tools you actually code with after the landing page is mocked.

Search "Cursor vs Claude Code vs OpenCode" and the top 10 is US-shaped: NxCode, ComputingForGeeks, ShareUHack, Respan, DevToolPicks. None of them are written from a UK indie hacker's chair, none of them quote GBP, and none of them tell you the honest answer — which is that most serious UK builders end up running two of these in tandem, not picking one.

This post is the UK-builder primer. GBP throughout. The three-question decision tree at the end. The two-tool combo most weekend builders settle on by month two.

The 30-second positioning

Tool	Shape	Who it's for
Cursor	VS Code fork, AI inside the editor	The builder who lives in their IDE and wants AI tab-complete + composer for refactors
Claude Code	Terminal-based agent (no IDE)	The builder who wants the AI to run an hour on its own — read files, edit, test, commit
OpenCode	Open-source terminal agent, bring your own LLM	The cost-sensitive builder + anyone who wants to pick their own model per task

These three are not direct competitors despite the SERP framing. Cursor is "AI in the IDE". Claude Code and OpenCode are "AI as the IDE". The choice between Cursor and the agents is a workflow question; the choice between Claude Code and OpenCode is a pricing-and-model-flexibility question.

The UK GBP cost table (May 2026)

Real monthly cost at the tier most UK weekend builders run at. All prices include VAT impact where relevant (USD-billed services are charged through the card's FX, so the GBP figure floats slightly).

Tool	Tier	Monthly cost	What you get
Cursor Pro	Pro	GBP 16-17	500 fast premium requests + unlimited slow, all models
Cursor Ultra	Ultra	GBP 160	20x more fast premium requests, priority support
Claude Code (Anthropic Pro)	Pro	GBP 17	Includes Claude Code with a soft quota, Sonnet/Opus access
Claude Code (Anthropic Max)	Max 5x	GBP 80	5x Pro quota, generous for serious vibe coding
Claude Code (Anthropic Max)	Max 20x	GBP 160	20x Pro quota, near-unlimited for solo builders
OpenCode	Free + BYOK	GBP 0 + LLM cost	Free tier, you pay your own API key (Anthropic, OpenAI, Groq, etc.)
OpenCode Go	Paid	GBP 8	Lowest-cost paid tier, no BYOK required for hosted models
OpenCode + Anthropic API	BYOK	~GBP 30-80/mo	At typical UK weekend usage, after the API top-ups

A note on the OpenCode + API maths: the model fees do not actually beat Claude Code Pro at GBP 17/month for typical UK indie hacker usage. OpenCode wins on flexibility (model routing per task) and transparency (you can see every token), not raw cost.

The decision tree

Three questions, in order. Answer them honestly.

Question 1: Do you live in an IDE all day?

If yes, Cursor is the natural fit. VS Code fork means every plugin you already know works. Composer mode lets you do agentic refactors without leaving the editor. The single biggest reason a builder regrets switching off Cursor is muscle memory — keyboard shortcuts, file tree mental model, debugger integration.

If no — if you spend equal time in the terminal, or you're a "scaffold and inspect" rather than "tab through 50 files" builder — skip to question 2.

Question 2: Do you want the AI to run for an hour on its own?

If yes, Claude Code is the right tool. The killer pattern is "vibe coding with subagents" — you give it a brief, it dispatches subagents, they run in parallel, they come back with edits, the main thread reviews and commits. That pattern is awkward in Cursor; it is native in Claude Code.

If you also want fine-grained model control — different model for research, different model for code-writing, different model for tests — go to question 3.

Question 3: Are you price-sensitive at the LLM-cost layer?

If yes, and you already have an Anthropic or OpenAI API key with credit, OpenCode lets you keep your existing spend and route per task. The Go plan at GBP 8/month removes the BYOK requirement if you'd prefer a hosted model.

If no — you'd rather have one bill at the end of the month — Claude Code on Anthropic Pro or Max is the cleanest.

The two-tool combo most UK weekend builders settle on

After about a month of solo building, most UK indie hackers we hear from converge on the same pattern:

• Cursor for IDE work — tab-complete in the day-to-day, composer for medium refactors, multi-file edits where you want to see every change inline.
• Claude Code for the big jobs — full feature scaffolds, "spin up a new microservice", overnight refactors that touch 40 files, exploratory research with subagents.

That combo is roughly GBP 33/month (Cursor Pro + Anthropic Pro). It's still well below a single freelance dev day (GBP 400-600), and the two tools complement each other rather than overlap. The reason it converges is that the things Cursor is brilliant at (in-IDE editing) are not the things Claude Code is brilliant at (long-running autonomous runs), and vice versa.

The three honest weekend scenarios

Scenario A: ship a paid micro SaaS in a weekend (Friday-Sunday)

Best tool: Claude Code. You want the agent to scaffold Next.js + Supabase + Stripe in one prompt, deploy to Vercel, and only come back to you when something needs human judgement. The IDE-bound shape of Cursor slows this down; the constant context-switching of OpenCode adds friction.

Scenario B: refactor a 6-month-old side project that's gone sideways

Best tool: Cursor. You want to see every change inline, you want to keep the diff small, you want to step through the affected files. Composer mode + Cursor's diff view shines here. Claude Code is overkill; OpenCode lacks the IDE muscle.

Scenario C: a research-heavy build with lots of "find me 10 examples of X" work

Best tool: OpenCode (or Claude Code if you don't mind the cost). The model-routing flexibility means you can send "find examples" to a cheap fast model (Haiku, Groq) and "write the actual code" to a strong slow model (Opus, GPT-5). That cost split is impossible inside Cursor and harder to set up inside Claude Code.

Who shouldn't use which tool

Don't use Cursor if you keep getting frustrated that the AI does not run for long enough on its own. Cursor's design philosophy is "human in the loop on every change". If you want agentic autonomy, that fight is going to wear you down.

Don't use Claude Code if you have a small project where one file at a time is the right shape. The agent loop overhead is not worth it for "fix this CSS bug in 5 minutes". Use Cursor for those.

Don't use OpenCode if you don't have an LLM API key already and you don't want to manage one. The BYOK setup adds real friction, and the Go plan (GBP 8) is a small saving over Claude Code Pro (GBP 17) that comes with much less polish.

What about VS Code + Copilot, Windsurf, Zed, and the rest?

Outside scope for today. The three tools above represent the three shapes (IDE-AI, agent-AI, open-source agent-AI) that most serious UK indie hackers are running in 2026. Windsurf is a Cursor-shaped competitor; Zed has a strong following but a smaller AI surface; Copilot is fine for autocomplete but not enough for vibe coding. If you have one of those already and it works for you, keep it — switching for the sake of it is a tax.

The cost ceiling worth knowing about

If you run two of these tools and your weekend project ships to paying customers, the right monthly stack tops out at:

• Cursor Pro: GBP 16
• Anthropic Max 5x (Claude Code): GBP 80
• (Optional) OpenCode + GBP 20-30 of API credit for cost-routing experiments
• Total: ~GBP 100-130/month

A single working weekend at that stack should produce more value than the monthly cost. If it doesn't, the problem is not the tools — it's the project. Cut the project, not the stack.

Want a data-backed UK business idea every week? Free reports drop every Thursday — keyword volumes, SERP analysis, builder prompts. Browse the latest free report on IdeaStack.

---