AI coding usage limits compared (and free alternatives when you run out)
What each plan actually allows, when the limit resets, and free or cheap fallbacks when you run out. Checked against vendor pages on 2026-10-20.
If you vibe code on X, you’ve seen the pattern: run out of Claude Code, switch to Codex, run out of that, then burn through Cursor credits. This page lists what each popular tool’s plans include, when the limit resets, and what to do when you hit the wall, with free and cheap fallbacks at the end. Every figure comes from the vendor’s own pricing or docs page, checked on 20 October 2026. Where a vendor doesn’t publish a number, we say so.
The short version
- Time-window limits (Claude Code, Codex, Windsurf/Devin Desktop, Cursor): you get a bucket that refills on a clock. Claude Code and Codex Plus use a 5-hour window plus weekly limits, Windsurf refreshes daily and weekly, and Cursor weekly. If you run out, waiting genuinely works.
- Credit or token budgets (Lovable, Bolt, v0, Replit): you get a monthly allowance, sometimes with a small daily freebie on top. Waiting only helps with the daily part.
- Every tool sells overage, usually at API rates. That’s the expensive way out, so try the fallbacks below first.
Usage limits compared (checked 2026-10-20)
| Tool | Plans and prices | How the limit works and when it resets | When you run out |
|---|---|---|---|
| Claude Code | Not on Free. Pro $20/mo ($17/mo billed annually). Max from $100/mo, with 5x or 20x Pro’s usage. | A rolling 5-hour session limit plus a weekly limit, shared between the Claude apps and Claude Code. Anthropic publishes no fixed message count. | Wait for the reset, upgrade, or turn on usage credits (billed at API rates). Check what’s left with /status.Source: Claude pricing · Claude Code help |
| OpenAI Codex | Free $0 and Go $8/mo (both list GPT-6 Luna in the desktop app, subject to rollout). Plus $20/mo. Pro $100, $200 or $500/mo. | Plus and Business are measured per 5-hour period, and weekly limits may also apply. OpenAI says Pro currently has no 5-hour limit. Reset times show in your usage dashboard. | Plus and Pro can buy extra credits. Anyone can run local chats with an API key at API rates. Check with /status.Source: Codex pricing |
| Cursor | Free $0 (limited). Plus $8/mo (Grok models only). Premium $30 (4x Plus). Super $100 (15x). Ultra $200 (50x). Teams from $30/user. | One usage pool shared across Cursor and Grok, which refreshes weekly. Third-party models are billed separately at API list price. | Buy top-offs or wait for the weekly refresh. Your remaining usage shows in settings, and you can set spending caps for third-party models. Source: Cursor pricing |
| Windsurf (now Devin Desktop) | Free $0 (light quota). Pro $20/mo. Max $200/mo. Teams $80/mo plus $40 per full dev seat. | A usage allowance that refreshes daily and weekly. Cost per message depends on the model and the size of the task. | Paid plans can buy extra usage at API pricing. windsurf.com/pricing now redirects to devin.ai/pricing. Source: Devin pricing · Cognition’s pricing announcement |
| Lovable | Free: 5 build credits a day, capped at 30 a month. Pro from $25/mo for 100 credits ($21/mo billed annually). Business from $50/mo. | Daily build credits and the daily chat allowance reset at 00:00 UTC. Free’s daily credits stop at the 30-a-month cap and resume on the 1st. Paid monthly credits arrive on your billing date and roll over, expiring two months after they’re issued. | Pro and Business can buy one-off or automatic top-ups, valid for 12 months. Plan mode costs 1 credit per message, so plan first and build second. Source: Lovable credits · Lovable plans |
| Bolt.new | Free: 1M tokens a month and 300K a day. Pro $18/mo billed annually ($216 a year), with 10M tokens a month. Teams $27 per member per month, billed annually. | Free has a daily cap and a monthly cap. Paid plans have no daily cap, and unused tokens roll over. Bolt doesn’t say what time the daily limit resets. | Upgrade, or keep projects small. Bolt says most tokens go on syncing your project’s files to the AI, so bigger projects cost more per message. Source: Bolt pricing |
| v0 by Vercel | Free: 7 messages a day. The paid tier is $30 per user per month, with $30 of monthly credits plus $2 of free credits each day you log in. Business $100 per user per month. | Free is capped per day. Paid plans spend credits by token at the rate of the model you pick (v0 Mini is cheapest). v0 doesn’t say what time the daily limit resets. | Buy extra credits, or switch to v0 Mini for routine edits. Source: v0 pricing |
| Replit | Starter (free) gets daily Agent credits up to a monthly cap. Core $20/mo ($18 billed annually), including “up to 30 hours of chat on Free Mode” and $20 towards more powerful models. The next tier is $100/mo ($90 billed annually). | Starter’s Agent credits reset daily, and its cloud credits reset monthly. Replit doesn’t publish exact Starter credit amounts on the pages we checked. | Upgrade to Core, which also adds pay-as-you-go. Source: Replit pricing · Starter plan docs |
Prices are USD and exclude tax and VAT. Usage-based limits change often, so check the linked page before you buy. None of these vendors publishes a fixed “messages per day” number for its paid agent plans. Codex and Windsurf give message ranges that depend on the model and the task.
Gotchas that eat your limit faster
- Claude Code and a stray API key: if
ANTHROPIC_API_KEYis set on your machine, Claude Code bills that key at API rates instead of using your Pro or Max plan. Run/logout, then log back in with your Claude account, and leave out any Console credentials if you want to stay on your plan. (Anthropic) - Chat and Claude Code share one pool: long chats in the Claude app count against the same 5-hour and weekly limits as your coding.
- Codex Fast and Ultrafast modes use your included usage 2.5x and 8x faster than Standard, and image generation uses it about 3–5x faster. (OpenAI)
- MCP servers and big instruction files: OpenAI says every MCP server you have switched on adds context to each message. It recommends switching off the ones you aren’t using and keeping
AGENTS.mdsmall. The same applies toCLAUDE.mdand Cursor rules. - Cursor third-party models: on Premium and above, non-Grok models are billed at API list price, separately from your plan pool. Set a spending cap.
- Bolt project size: Bolt says most token use comes from syncing your project’s files, so a large project costs more on every message.
Free and cheap fallbacks when you hit the wall
1. Free models on OpenRouter
OpenRouter lists models at $0 per million tokens, with IDs ending in :free. On 20 October 2026 these included MiniMax M3.1 Flash, NVIDIA Nemotron 3.5 Super VL and Lightning, Thinking Machines Inkling Small, and Dots3-Note Preview (which is going away on 31 December 2026). Free models are capped at 20 requests a minute and 50 a day until you’ve bought at least 10 credits in total. After that the cap rises to 1,000 a day. The daily count resets at 00:00 UTC. (OpenRouter limits)
Plug an OpenRouter key into any tool that accepts your own key or an OpenAI-compatible endpoint, such as Cline or a CLI agent. Free models come and go, and some providers say they keep prompts to train their models, so read each model’s page before pasting private code into it.
2. Google’s free tiers
Gemini CLI allows up to 1,000 model requests a day when you sign in with a Google account. With an unpaid Gemini API key you get the Flash model only, at 250 requests a day and 10 a minute. Google Antigravity, Google’s agent-first IDE, is free while it’s in public preview, with weekly rate limits. (Google Cloud) Both need a Google account.
3. Local models with Ollama
Ollama downloads open models and runs them on your own computer, and it can launch coding agents from your terminal. A local model has no usage cap: the limit is your hardware. Expect smaller local models to be noticeably weaker than frontier models at big multi-file changes. Use them for small edits, explanations and boilerplate.
4. The free tiers of the tools themselves
Every tool above except Claude Code has a free plan: Codex Free, Cursor Free, Windsurf/Devin Free, Lovable’s 5 daily credits, Bolt’s 300K daily tokens, v0’s 7 daily messages and Replit Starter. Spreading small tasks across them is a real strategy, but keep one tool as your main codebase owner so you don’t end up with four half-finished copies of your app.
How to make your limits last longer
- Plan cheaply, build expensively. Lovable’s Plan mode costs 1 credit per message. Do the thinking there, or in a free model, and only send the final, specific build prompt to the expensive agent.
- Use smaller models for routine work. Codex lists far more GPT-6 Luna messages than GPT-6.1 Sol per 5-hour window (350–3,000 against 15–160 on Plus). Devin says the same: switch to smaller models for routine work.
- Trim the context you send. Point the agent at the files that matter instead of the whole repo, and switch off MCP servers you aren’t using.
- Check before you start a big task. Run
/statusin Claude Code or Codex. In Cursor and Lovable, check the usage view in settings. - Know your reset clock. Lovable’s daily credits reset at 00:00 UTC, which is midnight in winter and 1am in summer for UK readers. Claude’s 5-hour window is rolling, and Codex shows your exact reset time in its usage dashboard. Check both before starting a long task.
Choosing a main tool? See Best AI code editors in 2026, Lovable vs Bolt vs Replit, and our coding agents ranking.
FAQ
When does the Claude Code limit reset?
Claude Pro and Max use a rolling five-hour session window plus a weekly limit, shared across Claude and Claude Code. Check where you stand with /status in Claude Code or under Settings > Usage. Anthropic doesn’t publish a fixed message count.
When does the Codex limit reset?
OpenAI measures Plus and Business usage per five-hour period, and weekly limits may also apply. Your exact reset time is in the Codex usage dashboard, or run /status in the CLI. OpenAI says Pro plans currently have no five-hour limit.
How often do Cursor credits refresh?
Cursor’s pricing page says each plan’s usage pool, shared across Cursor and Grok, refreshes weekly. When it runs out you can buy top-offs or wait. Third-party models are billed separately at API list price.
What’s the best free fallback when I run out?
Free models on OpenRouter (50 requests a day, or 1,000 a day once you’ve bought 10 credits), Gemini CLI (up to 1,000 requests a day with a Google account), or a local model via Ollama. Use these for small tasks and keep your main tool for big changes.
Do Lovable credits roll over?
Paid monthly plan credits roll over while your subscription is active, and expire two months after they’re issued (on monthly billing). The 5 daily build credits and the daily chat allowance don’t roll over. They reset at 00:00 UTC.