If you are choosing a coding agent this month, the price you see today may not be the price you pay in December. Google’s flagship model for its coding tools, Gemini 4 Argon, is on an introductory rate that doubles on 1 December 2026 — from $2 to $4 per million input tokens, and from $10 to $20 per million output tokens. That one change moves the maths for anyone comparing Codex, Claude Code and Google Antigravity, so this coding agent pricing comparison puts all three side by side on the numbers that actually decide your bill.
Disclosure: this article contains no affiliate links. Every link below goes to the vendor’s own official page, and no company paid for its place here.
How we compared: this is desk research, not a hands-on test — we have not run billed workloads on any of these tools. Every price below was checked on the vendor’s own pricing page on 5 November 2026, and the worked examples are our own arithmetic from those published rates, labelled as such. Prices are US list prices before tax, and vendors change them without much notice, so check the source link before you commit to a plan.
In this guide:
- Codex vs Claude Code vs Antigravity at a glance
- The 1 December change, explained
- Codex: what it costs
- Claude Code: what it costs
- Antigravity: what it costs
- What a real task costs on each one
- Which one fits which budget
- Frequently asked questions
Codex vs Claude Code vs Antigravity at a glance
| Codex (OpenAI) | Claude Code (Anthropic) | Antigravity (Google) | |
|---|---|---|---|
| Cheapest entry | $0 (Free trial) / $8 Go | $20/mo Pro | $0 Individual tier |
| Practical starting plan | Plus, $20/mo | Pro, $20/mo ($17 on annual) | Google AI Pro, $19.99/mo |
| Top individual plan | Pro 20x, $200/mo | Ultra, $200/mo | Google AI Ultra 20x, $199.99/mo |
| Pay-per-token route | API: $1.75 in / $14 out per 1M tokens (gpt-5.3-codex) | API: from $2 in / $10 out per 1M tokens (Sonnet 5.5) | Gemini API via CLI: $2 in / $10 out per 1M tokens (Argon, until 30 Nov) |
| How limits work | Credits; 5-hour windows and weekly limits | 5-hour session + weekly limits; Ultra drops the 5-hour limit | Quota by work done; weekly refresh free, 5-hour refresh paid |
| The catch | Output tokens cost 8x input; heavy refactors burn credits fast | No free tier includes Claude Code; top model costs extra on Pro | Quota is opaque; Argon burns it fast; no bring-your-own-key for extra limits |
The 1 December change, explained
The reason this comparison has a deadline is Gemini 4 Argon. Google launched the model on the Gemini API at an introductory price: $2 per million input tokens and $10 per million output tokens (thinking tokens included), with cached input at $0.10. That price runs through 30 November 2026. From 1 December, every rate doubles: $4 input, $20 output, $0.20 cached. Google has said nothing about extending it, so plan as if it ends on schedule.
Two details matter. First, the cheaper tiers double too: Batch and Flex, which cost half the Standard rate, go from $1/$5 to $2/$10 per million tokens — so Batch or Flex after 1 December costs exactly what Standard costs today. If your work can wait or tolerate retries, moving it off Standard before the deadline keeps that part of your bill flat. Second, Google’s pricing page does not state a time zone for the cutover, so do not schedule a large job for the last hours of 30 November and expect the old rate.
Argon is not the only Google model with a dated rise. Gemini 3.8 Flash, the model behind Antigravity’s free tier, doubles on 1 January 2027, from $0.75/$3.75 to $1.50/$7.50 per million tokens. And Argon has no free API tier at all: every token is billed, including thinking tokens and internal tool calls.
Codex: what it costs
Codex has no price of its own. OpenAI sells it inside ChatGPT, so what you pay for is the size of your allowance — plus, if you go the API route, the tokens you actually burn.
Subscription (ChatGPT plans, US): Free gives limited trial access at $0. Go is $8 a month for light, occasional use. Plus is $20 a month and is the practical default — it includes Codex on web, CLI, IDE extension and iOS. Pro split into two tiers in April 2026: Pro 5x at $100 a month and Pro 20x at $200 a month. Business seats are pay-as-you-go, and Enterprise is custom.
Since April 2026, every Codex plan runs on credits, and credits are just tokens in another costume: one credit works out to roughly $0.04 of API usage. On the current rate card, gpt-5.3-codex costs 43.75 credits per million input tokens, 4.375 for cached input and 350 for output — which is $1.75, $0.175 and $14.00 in dollars. Output costs eight times input, so a task that reads a lot and writes a little is cheap, and a long refactor that writes thousands of lines is not. OpenAI’s own estimate for active use is $100 to $200 per developer per month.
API (your own key, no subscription): gpt-5.3-codex bills at $1.75 per million input tokens, $0.175 cached and $14.00 per million output tokens, with no included allowance. The API route drops some conveniences — no automatic GitHub code review or Slack integration — and new models typically reach ChatGPT subscribers first. One caution: a pricing tracker flags gpt-5.3-codex as deprecated in OpenAI’s model catalog, so check which coding model your account can use before building on that rate. The newer comparable code model on the same rate card, GPT-6.1 Sol, is $2 input and $10 output per million tokens.
The honest rule for Codex is that the plan price never moves but the credits behind it do. A $20 Plus plan can feel generous one week and tight the next, depending entirely on how much output your tasks generate.
Claude Code: what it costs
Claude Code is the simplest of the three to price, because it is a flat subscription with no token meter — until you hit a limit. It is included in every paid Claude plan, and the Free plan does not include it at all.
Subscription (Anthropic, US, before tax): Pro is $20 a month, or $200 paid up front for a year, which Anthropic shows as $17 a month. Max is $100 a month and gives five times Pro’s usage every five hours and every week. Ultra is $200 a month, gives ten times Pro’s weekly usage, and drops the five-hour limit for verified accounts. Ultra replaced the former Max 20x tier at the same price. Team seats start at $25 a month ($20 on annual billing) for Standard and $125 ($100 annual) for Premium; Enterprise is $20 per seat per month on annual billing plus usage at API rates.
Two things are easy to miss. First, Max and Ultra include monthly API credits — $100 on Max, $200 on Ultra — but those credits do not pay for interactive Claude Code sessions or for usage credits inside your subscription; they are for separate scripted API work. Second, Claude Code defaults to Opus 5.5 on individual plans, and Anthropic’s top model, Fable, is included on Max and Ultra for up to half your weekly limit but costs extra usage credits on Pro.
API (pay per token, via Claude Console): if you would rather pay for what you use, the base rates are Sonnet 5.5 at $2 input and $10 output per million tokens, Opus 5.5 at $4 and $20, Haiku 5.5 at $0.10 and $0.50 (for prompts up to 100,000 tokens), and Fable 5.1 at $10 and $50. That makes Sonnet 5.5 the same price as Argon’s introductory rate — and the same price Argon moves to in December is Opus 5.5’s price today. For a fuller look at Claude’s plans, see our Claude review 2026.
Antigravity: what it costs
Antigravity is the odd one out: there is no Antigravity subscription. You download it free, and paid capacity comes bundled with a Google AI plan. That makes it the cheapest to try and the hardest to budget.
Individual tier, $0 a month: the agent with Gemini 3.8 Flash as the core model, unlimited Tab completions and Command requests, and a basic weekly agent quota that refreshes once a week. When it is gone, you wait for next week.
Google AI Pro, $19.99 a month: a high, generous quota that refreshes every five hours until you reach the weekly limit, a higher weekly limit, and — since 13 October 2026 — Gemini 4 Argon, but only on a paid Pro subscription, not a trial. Pro also unlocks Claude Opus 5.5 and Sonnet 5.5 inside Antigravity. In the Gemini app, Argon is rolling out to Pro and Ultra subscribers.
Google AI Ultra, $99.99 (5x) or $199.99 (20x) a month: the highest quota, refreshed every five hours, with the highest weekly limits and the same Argon access. Past the quota, Pro and Ultra can buy AI credits through Google One and spend them on overages; the overage setting is either Never (wait for the refresh) or Always (spend credits automatically).
Three caveats. First, the quota is opaque: Google says rate limits correlate with the amount of work the agent does, so a quick fix costs little and a long multi-step task costs a lot — and one report had a single Argon benchmark task using two-thirds of a weekly Argon quota. Second, Argon is available in the Antigravity App, CLI and IDE extensions, but not in the Antigravity IDE itself. Third, there is no bring-your-own-key option for extra rate limits on consumer plans; the CLI can run against the Gemini API with your own key instead, and then every token bills at Gemini API prices — which is where the 1 December doubling bites.
What a real task costs on each one
Subscriptions and token prices are hard to compare directly, so here is one identical job priced three ways. The job: one million input tokens (a large codebase context) and 100,000 output tokens (the code and explanation written back). These are our own calculations from the published rates, not measured bills — a real agent run makes many model calls, and caching changes the totals.
| Route | Now (to 30 Nov 2026) | From 1 Dec 2026 |
|---|---|---|
| Gemini 4 Argon API (Standard) | $3.00 | $6.00 |
| Codex API (gpt-5.3-codex) | $3.15 | $3.15 (no change announced) |
| Claude Code API (Sonnet 5.5) | $3.00 | $3.00 (no change announced) |
| Claude Code API (Opus 5.5) | $6.00 | $6.00 (no change announced) |
| Claude Code Pro subscription | $20/mo flat, within limits | $20/mo flat, within limits |
| Antigravity on Google AI Pro | $19.99/mo flat, within quota | $19.99/mo flat, within quota |
Two things jump out. First, at API rates the three are within cents of each other for this job — until December, when Argon doubles and becomes the most expensive of the three. Second, the subscriptions are the better deal for steady work: if you run more than about six or seven jobs like this a month, a $20 flat plan beats paying per token, as long as you stay inside its limits. The per-token routes win for occasional use and for automated jobs, where a flat plan would sit idle most of the month. And caching is the biggest lever nobody budgets for: repeated context on Argon costs $0.10 per million tokens instead of $2.00 — but Argon only supports implicit caching, which Google controls, so you cannot force a cache hit the way you can mark breakpoints in Claude.
Which one fits which budget
Just trying a coding agent, $0: Antigravity’s Individual tier is the only genuinely free starting point of the three, with a weekly quota and no card. Codex’s Free tier is a limited trial. Claude Code has no free route — Pro at $20 is the entry point.
One person building regularly, about $20 a month: this is the closest call in the comparison. Codex Plus ($20) and Claude Code Pro ($20, or $17 on annual) cost the same; Antigravity on Google AI Pro ($19.99) is a cent cheaper and also buys you Gemini in the Gemini app, 2 TB of Google storage and the rest of the Google AI bundle. Pick on workflow, not price: Codex for background agent work across repos, Claude Code for interactive terminal sessions, Antigravity if you already live in Google’s tools and want several models in one place.
Heavy daily use, $100–$200 a month: Claude Code Max ($100) and Ultra ($200) are the most predictable — flat fees with stated multiples of Pro’s usage. Codex Pro 5x and 20x match them dollar for dollar, but your real cost depends on credit burn, which varies with task size. Antigravity Ultra ($99.99/$199.99) is the least predictable of the three, because the quota is measured in work done, not tokens, and Google says limits can change to manage capacity.
Automated or occasional use: pay per token. Codex and Claude Code both bill cleanly by the token, and the Gemini API does the same for Argon — but Argon is the one to watch: after 1 December it costs the same as Claude’s Opus 5.5, the most expensive rate in this comparison, and its Batch/Flex half-price tiers become the way to keep December bills at November levels.
The bottom line: nothing about the 1 December rise makes Argon a bad model — it makes it a December decision. If you are already building on Argon, the useful moves before 30 November are to shift latency-tolerant work to Batch or Flex, lower the thinking level where quality allows, and check whether a $19.99 Pro subscription covers your volume more cheaply than the API. If you are choosing between the three agents rather than the three models, the $20 tier is effectively a tie, and the deciding factor is which workflow you will actually use every day. For broader context on AI agents, see our guide to Muse AI Agent, and our roundup of the best AI writing tools for bloggers.
Frequently asked questions
How much does Gemini 4 Argon cost after 1 December 2026?
On the Gemini API Standard tier, $4 per million input tokens and $20 per million output tokens, with cached input at $0.20 — exactly double the introductory rates of $2, $10 and $0.10 that run through 30 November 2026. Batch and Flex are half the Standard rate, so $2/$10 from December. There is no free tier for Argon.
Is Claude Code free?
No. The Free Claude plan does not include Claude Code. The cheapest way in is Pro at $20 a month, or $200 a year paid up front ($17 a month equivalent). The alternative is pay-per-token API billing through a Claude Console account, with Sonnet 5.5 at $2 input and $10 output per million tokens.
Is Google Antigravity really free?
The Individual tier costs $0 and includes the agent with Gemini 3.8 Flash, unlimited Tab completions and a basic weekly quota. Higher limits and Gemini 4 Argon need a Google AI plan: Pro at $19.99 a month, or Ultra at $99.99 (5x) or $199.99 (20x). Argon on Pro needs a paid subscription, not a trial.
Which coding agent is cheapest for a small team?
For one person, all three start at about $20 a month, so the price is a tie and the workflow decides. For a team, compare seat pricing rather than individual plans: Claude Team Standard seats are $25 a month ($20 annual), Codex Business seats are pay-as-you-go, and Antigravity for organisations runs through Gemini Enterprise, priced per seat or by consumption — get a quote for your team size rather than multiplying an individual plan.
Should I switch coding agents before 1 December?
Only if you are paying per token for Argon and your volume is high. The practical moves are cheaper than switching: move latency-tolerant Argon work to Batch or Flex, lower the thinking level, and compare a $19.99 Google AI Pro subscription against your API bill. If you use Claude Code or Codex on a flat subscription, the December change does not touch your bill at all.
Sources: prices checked 5 November 2026 on the vendors’ own pages — Gemini Developer API pricing (Gemini 4 Argon and Gemini 3.8 Flash rates); Claude plans and pricing (Pro, Max, Ultra and Team); Google Antigravity plans and Antigravity pricing (quota rules and Google AI plan prices); OpenAI ChatGPT rate card (Codex credit rates, including gpt-5.3-codex and GPT-6.1 Sol); Google’s Gemini 4 Argon announcement. Worked examples are our own arithmetic from those published rates. Prices are US list prices before tax and can change; check the source page before you buy.