Which AI coding tool is cheapest? Claude Code, Codex, Gemini CLI, and Cursor on a real monthly bill

The cheapest AI coding tool is the plan your month actually stays inside. Claude Code, Codex, Gemini CLI, and Cursor can sit at $0, $8, $20, or $100. After the allowance, Claude and Cursor bill overage in dollars per million tokens, Codex bills credits, and a personal Gemini terminal now runs through Antigravity CLI, which does not publish how many requests that allowance is. The four “included usage” units do not add up to one monthly bill.

Today is 20 October 2026. Claude’s seat prices and token rates are the ones checked yesterday on claude.com/pricing and the 7 October rate card. The full rate table is in yesterday’s note. Codex tiers, the five-hour estimates, and credit rates are from the Codex pricing page opened today. Cursor is from Models & Pricing and cursor.com/pricing. Gemini CLI signed in with a Google account — the free tier, Google AI Pro, and Google AI Ultra — stopped being served on 18 June 2026. See the migration note and the deprecation page. The personal terminal path is Antigravity CLI. Google AI prices are from the Google AI plans page and the Gemini subscriptions page. Gemini 4 Argon API rates are the ones checked in the 14 October note: standard tier is $2 input and $10 output per million tokens through 30 November, then $4 and $20 from 1 December.

The seat that actually stops

Each product has a seat fee and a different overage switch. The month collapses to a single number when you decline overage and the work is still inside the allowance.

• Claude Code. A Pro or Max seat covers the web app, the desktop app, and Claude Code, on one five-hour window and one weekly limit. Inside that allowance there is no per-token charge. Usage credits bill at API list price. The API credits bundled with Max pay for Claude Platform. Claude Code overage needs its own usage credits. With ANTHROPIC_API_KEY set, every turn bills at API list price from the first one, and the subscription allowance is unused.
• Codex. It rides on the ChatGPT seat and shares usage with ChatGPT Work. Plus publishes a range of local messages per five hours. Pro currently has no five-hour cap, and a weekly limit can still apply. Past the limit you can buy credits. An API key bills at API list price from the first turn and has no Cloud features. The pricing page says API token prices meter API-key usage. Using them to estimate tasks inside the included allowance will not match the bill.
• Gemini CLI. Google-account sign-in for the free tier, AI Pro, and AI Ultra stopped being served on 18 June. A personal account that wants Gemini in the terminal opens Antigravity CLI. There is no separate coding subscription. Without Google AI Pro or Ultra the quota refreshes weekly. Pro and Ultra refresh every five hours until a weekly cap. Google does not publish how many requests that quota is. Gemini CLI still serves paid API keys, and Code Assist Standard / Enterprise seats for organizations.
• Cursor. Paid plans include a weekly pool for Grok 4.7, Grok 4.6, Grok 4.5, and Composer 2.5, shared with Grok Chat, Imagine, Voice, and Bot. Plus is those models only. On Premium and above, Claude, GPT, and Gemini bill at API list price from the first request, paid from prepaid credit, outside the $30, $100, or $200 seat.

The four price sheets on 20 October

Claude Code

Individual plans, tax extra. Free has no Claude Code. Max is monthly only. Pro billed annually is $200, which is $17 a month. At the limit you can wait for the reset, or turn on usage credits.

Global list price per million tokens on overage: Haiku 5.5 is $0.10 input and $0.50 output when the prompt stays at or under 100,000 tokens; Sonnet 5.5 is $2 and $10; Opus 5.5 is $4 and $20; Fable 5.1 is $10 and $50. On Pro, Fable bills at list price and is outside the included allowance. US-only inference is 1.1× those input and output rates. Cache-write and cache-hit rates are in yesterday’s table.

Codex

Codex follows the ChatGPT seat. The table is today’s individual cards plus the API key. The three Pro prices match About ChatGPT Pro tiers: $100, $200, and $500, with Ultrafast only on the $500 plan. New subscriptions that are not grandfathered include a lower allowance. Pro 200 includes more than Pro 100. Pro 500 is the highest of the three.

Local-message estimates for Plus and standard Business are per five hours. The pricing page says they are not fixed caps. Cloud tasks spend more of the allowance. Local messages and cloud tasks share one allowance, and a weekly limit may also apply. Pro currently has no five-hour table.

After the included allowance, Standard-speed credit rates are per million tokens. Codex credit billing has no separate cache-write charge. GPT-6.1 Sol is 50 input, 2.5 cached input, and 250 output. GPT-6 Luna is 2.5 input, 0.25 cached input, and 12.5 output. Fast counts as 2.5× inside the included allowance and 2× on purchased credits. Ultrafast for GPT-6.1 Sol counts as 8× inside the allowance and 6× on credits. The help center says purchase amounts, prices, and payment methods vary by account, region, and plan. This note does not convert a credit into one universal dollar price.

Gemini CLI and Antigravity CLI

If the Gemini CLI instructions you are holding predate June, the sign-in has changed. The top of the Gemini CLI quota page says unpaid and Google One users moved to Antigravity CLI on 18 June. The page’s own last-updated line is 18 June 2026, and the body still prints the old personal rows of 1,000, 1,500, and 2,000 requests a day. The deprecation page says those tiers can no longer use Login with Google. For a bill you can pay today, use the deprecation page and the Antigravity plans page.

Google AI Plus is $4.99 a month. The Gemini CLI quota page says tiers that are not listed, including AI Plus, are unsupported. The Antigravity plans page puts the higher coding quota on Pro and Ultra. An account without Pro or Ultra keeps the weekly quota. The $4.99 buys other Google AI benefits. This note does not treat it as a coding-quota upgrade.

The Pro and Ultra overage switch is AI Credit Overages. Never means the model stops until the baseline quota refreshes. Always means purchased AI credits are used automatically, then the baseline quota resumes when it refreshes. Credits bill at Gemini Enterprise consumption rates. The Antigravity plans page does not publish a per-million-token table that this note can multiply. Remaining quota is /usage in Antigravity CLI.

Cursor

Cursor is sold as part of The X Subscription: one subscription for Grok, Cursor, and X. The table is today’s individual plans. Annual billing is $80 for Plus, $300 for Premium, $1,000 for Super, and $2,000 for Ultra, tax extra. The pool refreshes weekly, so one heavy day does not wipe out the month, but Grok chat and coding draw from the same pool.

After the weekly pool, Grok and Composer can be topped up at the extra-usage rates in the docs. Grok 4.7 is $2 input, $0.50 cache read, and $6 output per million tokens. Composer 2.5 is $0.50 input, $0.20 cache read, and $2.50 output. Claude, GPT, and Gemini are outside that pool. Individual plans do not pay the Cursor Token Rate. Teams and Enterprise add $0.25 per million tokens on third-party models. The personal totals in this note leave that line out.

What one overage turn costs

The same assumption lines up the overage each vendor actually prices: 80,000 input tokens, 2,000 billed output tokens, cache miss. It is not an average turn. Argon bills thinking tokens as output, so the 2,000 here already includes thinking. If the visible answer is 2,000 and thinking is extra, the Argon line is higher. The Codex lines are in credits. The help center does not publish one dollar conversion.

• Claude Sonnet 5.5, usage credits. 80,000 / 1,000,000 × 2 + 2,000 / 1,000,000 × 10 = $0.18.
• Claude Haiku 5.5, prompt still at or under 100,000. 80,000 / 1,000,000 × 0.10 + 2,000 / 1,000,000 × 0.50 = $0.009.
• Grok 4.7 bought past the Cursor weekly pool. 80,000 / 1,000,000 × 2 + 2,000 / 1,000,000 × 6 = $0.172.
• Composer 2.5 bought past the pool. 80,000 / 1,000,000 × 0.50 + 2,000 / 1,000,000 × 2.50 = $0.045.
• Gemini 4 Argon API, standard tier, through 30 November. The same token counts are $0.18. From 1 December, input $4 and output $20, the turn is 80,000 / 1,000,000 × 4 + 2,000 / 1,000,000 × 20 = $0.36.
• Codex, GPT-6.1 Sol, Standard, no cache. 80,000 / 1,000,000 × 50 + 2,000 / 1,000,000 × 250 = 4.5 credits.
• Codex, GPT-6 Luna, Standard, no cache. 80,000 / 1,000,000 × 2.5 + 2,000 / 1,000,000 × 12.5 = 0.225 credits.

Those seven lines appear only on the matching overage path. A Claude turn that is still inside Pro or Max has $0 of overage. Grok or Composer still inside the Cursor Plus weekly pool does not pick up the top-up rate. Codex credits are spent after the included allowance. Antigravity’s included quota has no unit-price line to put here.

The part of a month you can calculate

The 800 turns below are an assumption: 40 turns on each of 20 working days, still 80,000 input and 2,000 billed output, and every turn sits outside the included allowance. They are not a usage study. If the 800 turns stay inside the plan, overage is $0 and the month returns to the seat.

Add the seat, and keep the assumption that all 800 turns are overage:

• Stay inside the allowance. Cursor Plus is $8, about $6.67 a month on the annual bill. ChatGPT Go is $8. Claude Pro is $20, or $17 on the annual bill. Codex Plus is $20. Google AI Pro is $19.99, and that price also buys storage and other Gemini benefits. Coding is one of them. Antigravity without Pro or Ultra is $0 until the unpublished weekly quota stops.
• Claude Pro plus cold Sonnet overage. 20 + 144 = $164. The annual seat is $17 and the overage rate is unchanged, so the total is $161. Short-prompt Haiku is 20 + 7.20 = $27.20.
• Cursor Plus plus Grok 4.7 overage. 8 + 137.60 = $145.60. Composer 2.5 is 8 + 36 = $44. Plus cannot move those 800 turns onto Claude or GPT.
• Sonnet 5.5 inside Cursor Premium from the first request. 30 + 144 = $174. The Premium seat does not include Claude tokens. If the same 800 turns still fit in Claude Pro, Claude Code is $20, while Cursor meters list price from the first request.
• Codex Plus plus 3,600 GPT-6.1 Sol credits. The $20 seat is separate from 3,600 credits. OpenAI does not publish one dollar price that fits every account. Checkout is Settings → Usage, or Usage & Billing in Codex. The low end of the Plus Sol estimate is 15 local messages per five hours, so 40 turns in a day can leave that window. The high end is 160. Do not write Plus down as already covering these 800 Sol turns.
• Argon on a paid API key. Through 30 November that is $144 of tokens and no second seat. The same assumption from 1 December is $288. The $19.99 Google AI Pro plan does not include the Argon API.

Pick the cheap line for the job

• Grok or Composer is enough, and the week stays in the pool. Cursor Plus at $8 is the lowest published individual seat that includes a coding agent. Annual billing is about $6.67 a month. Chat in the Grok apps spends the same weekly pool.
• Lightweight GPT-6 Luna only. ChatGPT Go is $8. The pricing page puts web, CLI, IDE, iOS, and Cloud on the $20 Plus card. GPT-6.1 Sol, or Cloud code review, is the step up to Plus.
• A few focused GPT-6.1 Sol coding sessions a week. Codex Plus is $20. The pricing page’s own line is “a few focused coding sessions each week.” If the low end of the five-hour window is a regular event, read the usage dashboard before buying credits or moving to Pro, which currently has no five-hour cap.
• Claude in the terminal, and the work stays in the window. Claude Pro is $20, or $17 on the annual bill. Do not also select Claude inside Cursor Premium for the same repository. The Claude Code extension draws the subscription allowance. Claude selected inside Cursor bills at API list price from the first request.
• You already pay for Google AI Pro. Use Antigravity CLI. Inside the quota the marginal coding cost is $0. /usage shows what is left. Gemini CLI Google sign-in will not serve that subscription. Scripts or CI that need Gemini 4 Argon use a paid API key, at $2 / $10 per million tokens, then $4 / $20 from 1 December.
• One afternoon spends the five-hour window. The window rolls. Read Claude’s /status, the Codex usage dashboard, Antigravity’s /usage, and Cursor settings. The weekly limit, or Cursor’s weekly pool, is what says the month is short. Then compare the next seat with the overage rate. Eight hundred cold Sonnet turns are $144 of overage, so Pro at $20 plus that $144 is $164. Max 5x at $100 is lower only when those 800 turns actually fit in the Max window. Anthropic does not publish that token count.
• One repository, one primary seat. Claude Pro at $20 plus Cursor Premium at $30 is $50 of two seats. Claude used inside Cursor is metered from the first request on top of that. An API key on Codex leaves the ChatGPT allowance unused. For a hard ceiling, turn off automatic top-up: decline Claude usage credits and unset the API key, set Antigravity overage to Never, and set a monthly cap on Cursor additional usage.

Which line is the Mac

None of the numbers above include a machine. Add a Mac when the command has to be xcodebuild, the simulator, or signing. On the plan page today, M4 · 16GB / 256GB starts at $101.7 a month, and 24GB / 512GB starts at $200.3 a month. Checkout is the plan page. Claude Max 5x at $100 plus the 16GB line at $101.7 is $201.7 on two bills. None of the four coding subscriptions includes the machine.

Agent tool calls and the cloud server are a separate line. See the agent monthly-cost note. Do not record the same machine once under the coding subscription and again under the server bill.

FAQ

Is Cursor Plus at $8 the cheapest?

When Grok and Composer are enough and the weekly pool is not spent, $8 is the lowest published individual seat, about $6.67 a month on the annual bill. ChatGPT Go is also $8, and the model on that card is GPT-6 Luna. Antigravity CLI without Google AI Pro or Ultra has a $0 seat and an unpublished weekly quota. When that quota stops, you cannot continue with your own API key. Claude, GPT-6.1 Sol, and Gemini 4 Argon are not included in the $8 tier.

Can Gemini CLI still sign in with a Google account?

The personal free tier, Google AI Pro, and Google AI Ultra lost Login with Google on 18 June 2026. The terminal path is Antigravity CLI. A paid Gemini API key can still use Gemini CLI. Code Assist Standard and Enterprise organization seats still use Gemini CLI, at 1,500 or 2,000 model requests a day, and one prompt can become several requests. The old daily caps in the quota-page body were last updated on 18 June and conflict with the shutdown notice at the top of that page. Use the deprecation page.

How many dollars is Codex after the allowance?

Inside the allowance the month is the seat: Go $8, Plus $20, Pro $100, $200, or $500. Past the allowance, usage is credits. In this note’s example, 800 cold GPT-6.1 Sol turns are 3,600 credits, and GPT-6 Luna is 180 credits. OpenAI says the purchase price varies by account, region, and plan. There is no single dollar rate to multiply into every bill. An API key bills at API list price from the first turn and does not use the subscription allowance.

Is Claude inside Cursor cheaper than Claude Code on its own?

Changing the shell does not lower the list price. Cursor Plus cannot select Claude. Premium is $30, and Claude bills at API list price from the first request. One cold Sonnet 5.5 overage turn is $0.18, 800 turns are $144, and the Premium seat makes $174. If those 800 turns still fit in Claude Pro, Claude Code is $20. Cursor’s weekly pool covers Grok and Composer.

Do these four subscriptions include a Mac for iOS builds?

They do not. The Claude, Codex, Antigravity, and Cursor seats do not include xcodebuild, the simulator, or signing. Add a Cloud Mac when the command reaches that step. The 16GB tier starts at $101.7 a month today, recorded separately from the coding subscription.

Keep one seat that can stop

Inside the allowance, Cursor Plus and ChatGPT Go are $8, Claude Pro and Codex Plus are $20, and Google AI Pro is $19.99. Personal Google sign-in for Gemini CLI stopped on 18 June. The personal terminal is Antigravity CLI. Outside the allowance, Sonnet 5.5 and standard Argon through 30 November are both $0.18 a turn in this example, a Grok 4.7 top-up is $0.172, Composer 2.5 is $0.045, and GPT-6.1 Sol is 4.5 credits. If all 800 turns are overage, Sonnet is $144 and Sol is 3,600 credits. The dollar price of a credit is the checkout price. Keep one primary seat per repository. Add a Mac only when the command cannot leave macOS.

Further reading

• What Claude Code costs in a month: the seat, the overage, and usage credits
• Gemini 4 Argon API: input and output token prices, and a monthly cost
• Is GPT-6 worth the upgrade? Free, Plus, Pro, and Intelligent UI
• What an AI agent costs per month: API tokens, tool calls, and the cloud server