This is the free-tier special of our hardcore comparison column. The rule is stated up front: we compare only what you get without spending money, the free plans, limited promotions, credit rules, and free models of each AI coding tool. We pass no judgment on which model is smarter. If you came for "whose model is best", read our Claude Code versus Cursor and Codex capability review instead.
1. Scope and Selection Criteria
To avoid overlapping with existing comparisons, the division is clear. The Qoder international versus Qoder CN review covers edition differences in models, workflow, and compliance. The Claude Code versus Cursor and Codex review covers capability and workflow of three terminal agents. This article counts only the free-credit ledger; the three pieces complement each other.
Three criteria apply. First, the tool must target individual developers and be mainstream; team and enterprise tiers are excluded. Second, a free entry point must exist for individuals, a plan, a promotion, or a daily reward; tools with none are named but not expanded. Third, every number has an official source, referenced on 2026-09-18; where unverified, we describe it qualitatively and point to the pricing page, never inventing figures.
Six products made the cut: Qoder, Cursor, Trae, Windsurf, Claude Code, and Codex. Qoder launched a limited-time free promotion on 2026-09-18; details are in our Qoder Qwen3.8-Flash free hotspot coverage. For a routine to spend the daily free quota fully, see our Qoder free-credit SOP.
2. Free Ledger, Vendor by Vendor
Qoder: A Limited Window Plus Daily Refills
Qoder's permanent free tier is not generous. The international free tier costs 0 dollars but ships very few or zero Credits, opening only basic completion. The domestic Qoder CN adds a permanent free community tier at 0 yuan with about 100 Credits per month for basic completion and simple Q and A, but no multi-file context or refactoring. In short, the permanent free tier is tryable, not usable.
The real dividend comes from two campaigns on 2026-09-18 (source: ITHome, 2026-09-18). First, free Qwen3.8-Flash calling: from 10:00 on September 18 through 23:59:59 on September 30, its billing coefficient drops from 0.1 to 0.0, so every call costs no Credits. Users select the model in the picker; no claim is needed. Second, a daily login reward: a new round opens at 10:00 daily and can be claimed until 10:00 the next day. Each account claims 100 universal Credits once per round, valid 30 days, and unexpired grants accumulate. The reward lands in Add-on Credits across desktop, IDE, JetBrains plugin, CLI, Cloud Agents, mobile, and web. Eligibility covers all individual users of both editions, including free and trial users.
Across the thirteen-day window, an individual receives a steady 100 Credits per day stacked with zero-cost Qwen3.8-Flash calling. After it closes, Qwen3.8-Flash resumes billing at the published coefficient, and the daily reward's future follows official announcements.
Cursor: Two Thousand Completions Plus Fifty Slow Requests
Cursor's free tier is Hobby (source: cursor.com/pricing, 2026-09-18). It includes 2,000 code completions per month and 50 slow premium-model requests, with slow models such as Claude Sonnet and the GPT family, plus basic codebase indexing and full editor features. New accounts get a 7-day Pro trial, then fall back to Hobby. No credit card is needed.
The 50 slow requests are the quota for top models in Chat, Composer, and Agent, but sit in a slow queue at peak hours. Two thousand completions burn fast and typically hit the ceiling within a week for full-time developers. Verdict: Hobby suits evaluation, not daily free work.
Trae: The Most Generous Free Tier
Trae (international, trae.ai/pricing, 2026-09-18) ships the most relaxed free tier here. It includes 10 fast requests plus 50 slow premium requests (GPT-5 family, Gemini, Kimi, MiniMax, Seed) per month, 1,000 advanced-model requests, and 5,000 autocompletions, no card required. Pro is 10 dollars per month (3 dollars first month, 7-day trial); Lite is 3 dollars per month.
The 10 fast requests are priority calls with under 15 seconds queue wait; the 50 slow requests queue; the 1,000 advanced requests cover the next model tier; the 5,000 completions are a full bundle. For light to moderate users this sustains daily coding; the real limit is the 10 fast-request cap, with SOLO Agent mode unavailable free. The domestic trae.cn differs (500 points per month free); we use international figures.
Windsurf: Twenty-Five Credits Plus Five Sessions
Windsurf (formerly Codeium, windsurf.com, 2026-09-18) switched from credits to quotas on March 19, 2026. The free tier costs 0 dollars and includes a light quota, SWE-1-mini, restricted Cascade access, and unlimited Tab completion. Concretely: 25 prompt credits per month plus 5 Cascade sessions per day, no priority support. Pro is 20 dollars per month after the March 2026 increase.
Twenty-five credits equal roughly 3 to 5 meaningful Cascade sessions, enough to evaluate, not for daily use. The bright spot is unlimited Tab completion at every tier, rare among Agent IDEs. The constraint is the daily Cascade cap and monthly ceiling; a few complex multi-file tasks drain it.
Claude Code: No Free Tier
Claude Code (Anthropic, anthropic.com/pricing, 2026-09-18) has the clearest conclusion: no free tier. The Free plan (0 dollars) covers chat only, not Claude Code access. To use it you need Pro (20 dollars per month), Max (100 or 200 dollars), a Team seat, or an API key for usage billing. The CLI installs free, but running it needs a paid account with model access or a funded Console account.
API billing is the only no-subscription path: bind your key and pay per token. Sonnet runs about 3 dollars per million input and 15 per million output; Haiku about 1 and 5 (2026-09-18, per official page). For users who want to try without spending, Claude Code is not a candidate. It is pay-first.
Codex: A Limited Free Trial
Codex (OpenAI, developers.openai.com/codex/pricing, 2026-09-18) has no standalone subscription; its quota bundles into ChatGPT plans. The Free plan (0 dollars) lets you explore Codex on quick coding tasks, a restricted trial, not daily use. Go (8 dollars) targets light coding. Plus (20 dollars) unlocks full Codex across CLI, IDE, web, and cloud review. Pro offers 5x or 20x the Plus quota at 100 and 200 dollars.
The Codex CLI is open source and free, but GPT access depends on a ChatGPT plan or your API key. The Free tier's quick-task exploration suits zero-cost sampling, not daily free use. For pure free riders, Codex's window is wider than Claude Code's but narrower than Trae's or Qoder's.
3. Master Comparison Table
All numbers reflect official figures on 2026-09-18; entries marked per official page mean no fixed value is published.
| Product | Free tier price | Core free allowance | Free models | Key limits | Source |
|---|---|---|---|---|---|
| Qoder | 0 (permanent community) | ~100 Credits/month; 100/day in window | Basic; Qwen3.8-Flash free in window | No multi-file context; window Sep 18 to Sep 30 | ITHome 2026-09-18 |
| Cursor | 0 (Hobby) | 2000 completions + 50 slow premium | Claude Sonnet, GPT (slow) | 50 in slow queue; falls back after trial | cursor.com/pricing |
| Trae | 0 | 10 fast + 50 slow + 1000 advanced + 5000 completions | GPT-5, Gemini, Kimi, Seed, etc. | 10 fast only; SOLO off | trae.ai/pricing |
| Windsurf | 0 | 25 prompt credits + 5 Cascade/day | SWE-1-mini | Daily Cascade cap; monthly ceiling | windsurf.com |
| Claude Code | No free tier | API or paid plan only | None | Free excludes access | anthropic.com/pricing |
| Codex | 0 (trial) | Quick task exploration | GPT family (capped) | Not daily; needs Plus | developers.openai.com/codex |
Mapped to user types:
| User type | Best free option | Reason |
|---|---|---|
| Pure free rider | Trae free tier | Widest free bundle |
| Window arbitrage | Qoder Sep 18 to Sep 30 | 100 Credits/day plus free Qwen3.8-Flash |
| Light daily | Cursor Hobby | 2000 completions evaluate, 50 try models |
| Try Agent | Windsurf free | 5 Cascade/day plus unlimited completion |
| Zero-cost taste | Codex Free | Quick exploration, no payment |
| No trial without pay | Claude Code | No free tier, API from zero |
4. Strategy: Free Rider, Light User, Heavy User
Free riders pay nothing and live on free tiers. Their optimal setup is Trae free tier as base, stacked with Qoder's window (Sep 18 to Sep 30) of 100 Credits per day and free Qwen3.8-Flash. The two complement: Trae covers daily completion and mid-tier requests, Qoder covers heavy models and Agent tasks. After the window, keep Qoder for light completion and move the load back to Trae. To maximize the Qoder window, claim daily on schedule per our Qoder free-credit SOP.
Light users code a few times a week on completion and Q and A. Cursor Hobby's 2,000 completions plus 50 slow requests are enough; for more model choice, Trae is roomier. They need not rush to subscribe. Run the free tier a week or two, watch consumption, then decide. It helps to consult the Qoder international versus CN review to pick a side.
Heavy users run long Agent sessions and multi-file refactors daily. Frankly, none sustain that: Cursor's 50 slow, Windsurf's 25 credits, and Trae's 10 fast exhaust in a day or two. Heavy users must subscribe (Cursor Pro 20, Trae Pro 10, Windsurf Pro 20, Qoder Pro 20, Codex Plus 20, Claude Code Pro 20 dollars) or use API, and the free tier only helps them evaluate and arbitrage.
5. Window Arbitrage: Harvesting Limited Promotions
Limited promotions are the most overlooked part of the free ledger. Qoder's 2026-09-18 campaign is textbook: free Qwen3.8-Flash through September 30, plus 100 stackable Credits per day valid 30 days. The logic treats the window as a free-credit reservoir: route every task Qwen3.8-Flash can handle to that model, reserve plan allowance and daily rewards for pricier models such as Claude or Opus. Before the window closes, run pre-payable tasks early, riding half a month of allowance in thirteen days.
Arbitrage has boundaries. First, campaigns can adjust or end early; Qoder stated that after the free period Qwen3.8-Flash resumes billing at the published coefficient, and the daily reward follows announcements. Second, cross-tool arbitrage carries migration cost: each vendor's model pool, context engineering, and Agent workflow differ, and switching burns energy. Third, prices likely rise or tighten after the window, and dependency becomes a liability. For details and deadlines, see our Qoder Qwen3.8-Flash free hotspot coverage.
6. Cold Thinking: The Real Cost of Free Tiers
Free tiers are never zero cost; the bill changes form.
The first bill is data. Free tiers often mean your code, context, and conversations become training or product material unless you opt out or upgrade to a paid tier promising not to train. Windsurf writes zero retention on paid tiers, not free; Claude Code's Free plan lacks access, ironically cleaner. Free and private tend to move opposite.
The second bill is lock-in. Credit and quota systems bind your workflow to one platform. Once projects, teams, and habits migrate, switching dwarfs the subscription fee. Qoder's Credits, Cursor's pool, and Windsurf's quota are all "the deeper you use, the harder you leave". The free tier is an entry, not a destination; its value is zero-cost selection, not long-term reliance.
The third bill is the post-window hike. Today's generous free tier is usually acquisition marketing. The pattern is consistent: give it free, then tighten or raise. Qoder CN ended a limited free period in May 2026; Windsurf raised Pro from 15 to 20 dollars in March 2026. The free ledger must count not only savings now but the cost to leave once the habit forms.
Closing: the conclusion is not "whose free tier smells best" but "treat the free tier as a selection tool, not a plan". Free riders and light users get by on Trae and the Qoder window; heavy users face the bill. The free tier is bait and touchstone. Use it to find the product like your work, then read the Claude Code versus Cursor and Codex capability review for real strength.
FAQ
question1: What does this comparison compare, and what does it skip?
A1: Only what you get without spending: free plans, promotions, credit rules, free models. It does not judge capability. For capability see claude-code-vs-cursor-vs-codex; for editions see qoder-vs-qoder-cn-review.
question2: Exactly when does Qoder's limited free period run?
A2: Free Qwen3.8-Flash calling runs from 10:00 on 2026-09-18 through 23:59:59 on 2026-09-30. The 100-Credit daily reward starts at 10:00 each day from 2026-09-18, valid 30 days, stackable. Its future follows official announcements. Source: ITHome 2026-09-18.
question3: Which of the six has the most generous free tier?
A3: On international figures, Trae is widest (10 fast + 50 slow + 1000 advanced + 5000 completions). Qoder follows during the Sep 18 to Sep 30 window with 100 Credits per day and free Qwen3.8-Flash. Claude Code has no free tier; Codex offers only a trial.
question4: Can I code daily without paying?
A4: Barely, and the experience is limited. The most viable combo is Trae free as base plus Qoder window benefits. Heavy Agent use or multi-file refactoring hits every free ceiling; long term you need a subscription or API.
question5: Will the free tier train on my code?
A5: It depends on each vendor's privacy terms. Free tiers usually do not guarantee "not used for training"; paid tiers promise zero retention or enterprise compliance. Read the official privacy notes. Free and private tend to move opposite.