Claude Code pricing looks simple until you line it up against Codex and Gemini CLI. One tool sells seats and plan tiers, one hides inside ChatGPT subscriptions, and one is free software that bills you through whichever model you plug in. As of October 25, 2026, Google lists the Gemini 4 Argon launch rate at $2 per million input tokens and $10 per million output tokens, and the same announcement says the price rises to $4 and $20 after November 30, according to Google’s Gemini 4 Argon announcement.
That single scheduled increase changes the math for anyone budgeting a team around Gemini CLI. Anthropic moved in the other direction on Claude Opus 5.5, which it says costs 40% less to run than Opus 5 on typical workloads. OpenAI added a $100 Pro tier on October 9 and a $500 Pro tier on September 29. Three vendors, three pricing philosophies, and five weeks of changes.
This guide compares what each agent costs, what the money buys, and where the published numbers stop. We covered answer quality in our earlier look at Gemini CLI versus Claude Code, so this piece sticks to dollars, limits, speed tiers and the billing traps that surprise teams after the first invoice. Every figure comes from vendor pages or release notes fetched on October 25, and anything we could only find on a single source carries a hedge.
Don't miss new tech stories on Google
Add Tech Insider once in the Google app and our stories appear in your news suggestions.
Claude Code Pricing vs Codex vs Gemini CLI: The Short Answer
Pick Claude Code if you want a flat monthly subscription with a clear upgrade path and you already trust Anthropic’s Opus 5.5 for hard refactors. Pick Codex if your team already pays for ChatGPT and wants coding agents bundled with chat, a desktop app and cloud sandboxes. Pick Gemini CLI if you want an Apache 2.0 licensed tool, pay-per-token control, and the lowest list price per million tokens before December 1.
The longer answer depends on how you work. A developer who runs a few focused sessions a day will almost always pay less on a subscription than on raw API tokens. A developer who runs agents in parallel all day can blow through a $200 plan and land in credit or API territory within a week. A team lead who needs predictable invoices should care more about seat pricing and pooled credits than about the cheapest token.
Here is the one-paragraph scorecard we will defend in the verdict section:
- Lowest entry cost for light use: Codex, because OpenAI says Free and Go accounts include limited Codex access.
- Clearest individual ladder: Claude Code, with Pro at $20 monthly and higher Max tiers above it.
- Lowest published token price today: Gemini 4 Argon at $2 and $10 per million tokens, but only through November 30.
- Fastest paid speed tier: Codex Ultrafast, which OpenAI says generates tokens up to eight times faster on Pro 500.
- Most open: Gemini CLI, licensed under Apache 2.0.
What Changed in October 2026
The pricing story moved fast, so a timeline helps. The table below lists the dated changes that affect what a developer pays or which model the agent runs. Dates come from vendor announcements and help pages.
| Date | Vendor | Change |
|---|---|---|
| September 22, 2026 | Anthropic | Claude Opus 5.5 announced for agentic coding, 40% cheaper to run than Opus 5 on typical workloads |
| September 28, 2026 | Anthropic | Claude Sonnet 5.5 released, 30%+ faster and up to 30% cheaper than Sonnet 5 |
| September 29, 2026 | OpenAI | Pro 500 launched at $500 per month, with Ultrafast access |
| October 7, 2026 | Anthropic | Claude Haiku 5.5 announced, available in Claude Code |
| October 9, 2026 | OpenAI | $100 Pro tier introduced, Codex usage rules revised for Plus and Pro |
| October 13, 2026 | Gemini 4 Argon opens to developers at $2 and $10 per million tokens | |
| October 2026 | Anthropic | Five-hour usage limits raised for Pro, Max, Team and seat-based Enterprise |
| October 22, 2026 | OpenAI | GPT-6.1 Sol formally launched for ChatGPT Work and Codex |
Two details deserve attention. First, OpenAI’s Pro $200 plan stayed at $200, but customers who held it during the late-September window keep their previous included allowance only through October 29. After that the allowance shrinks while the price holds. Second, Google’s introductory Argon pricing has a hard end date, so any budget you write this month should include a second column for December.
If you compare model tiers rather than tools, our breakdown of model-tier price gaps across Sonnet 5.5, GPT-6 Luna and Gemini 3.8 shows how the cheap tiers stack up before the newest releases.
Specs Compared: 12 Rows on Models, Surfaces and Speed
Before the money, the basics. The table compares what each product actually is on October 25. Where a cell says “not confirmed”, our fetched sources did not state the answer, and we would rather leave a gap than guess.
| Spec | Claude Code | OpenAI Codex | Gemini CLI |
|---|---|---|---|
| Maker | Anthropic | OpenAI | |
| Newest model named for the tool | Claude Opus 5.5 (Sonnet 5.5 and Haiku 5.5 also available) | GPT-6.1 Sol (alongside GPT-6 Sol and GPT-6 Luna) | Gemini 4 Argon is live on the paid Gemini API, default CLI model not confirmed |
| Newest model release date | Opus 5.5 on September 22, 2026 | GPT-6.1 Sol formally launched by October 22, 2026 | Gemini 4 Argon on October 13, 2026 |
| Cheapest model tier | Haiku 5.5 at $0.10 and $0.50 per million tokens (prompts up to 100,000 tokens) | GPT-6 Luna, positioned for cost-sensitive volume | Not confirmed for the CLI |
| Where it runs | Terminal and IDE | CLI, IDE extension, desktop app, web, mobile, Codex Cloud | Terminal, with Antigravity as Google’s separate agent platform |
| Cloud sandboxes | Not confirmed | Codex Cloud on Plus, all Pro tiers, Business, Enterprise, Healthcare, Education | Not confirmed |
| Free access | Haiku 5.5 selectable on the Claude Free plan in Claude.ai, Claude Code on Free not confirmed | Limited Codex on Free and Go | README advertises a free tier, wording looks stale |
| Entry paid individual plan | Pro, $20 monthly | Plus, price not confirmed in our sources | Pay per token on the paid Gemini API |
| Top individual plan | Max tiers above Pro | Pro 500 at $500 per month | Not confirmed for the CLI |
| Speed premium | Fast mode on Opus 5.5, up to 2.5x faster at 2x standard price | Ultrafast, up to 8x faster token generation | None confirmed |
| Cache and batch discounts | Batch processing saves 50%, prompt caching up to 90% | Not confirmed | Cached input 95% cheaper on Argon |
| License | Not confirmed as open source | Not confirmed as open source | Apache 2.0 |
Three rows matter most for buyers. The speed premium row shows OpenAI and Anthropic both charging for latency, while Google has not announced a CLI equivalent. The license row shows why platform teams that need to audit or fork their tooling look at Gemini CLI first. And the cheapest-tier row shows Anthropic is the only vendor that publishes a token price for its small model in the sources we could verify.
Claude Code Pricing: Plans, Seats and Token Rates
Claude Code does not have a single sticker price. You reach it through a Claude subscription, through a Claude Console account that bills API tokens, or through a team seat. Anthropic’s Claude Code documentation describes the product as a coding agent for the terminal and IDE, and the cost depends on which of those routes you choose.
Subscription plans
Anthropic’s pricing page shows the first paid individual plan at $20 per month when billed monthly, or $200 billed up front on the annual option. A DEV Community write-up on Claude Code pricing frames the ladder as $20 for Pro and $100 for Max, with an API average of $13 a day in its headline. Treat that last figure as one author’s estimate, not an Anthropic benchmark.
For teams, the pricing page lists seat prices of $20 per seat per month on annual billing and $25 on monthly billing, plus a higher seat that lists $125 per month on monthly billing. Anthropic says Claude Code comes with every seat, with self-serve seat management and up to $500 per month in pooled API credits. The page also references weekly limits on premium seats, which signals that Anthropic meters heavier users separately from standard ones.
Anthropic also raised five-hour usage limits for Pro, Max, Team and seat-based Enterprise plans this month and added a user-selectable rate-limit reset. That change matters more than any price cut, because limits decide whether a flat plan holds up on a heavy week. For a task-level view of how those limits translate into cost, see our Claude Code versus Codex cost comparison.
API token rates
On the API side, the Fast mode page prices Opus 5.5 Fast at $8 per million input tokens and $40 per million output tokens, and says that is double standard pricing. Anthropic’s Opus 5.5 material lists standard rates of $4 and $20, with cache reads at $0.20 per million tokens, though we found that on one source only. Haiku 5.5 costs $0.10 input and $0.50 output for prompts up to 100,000 tokens, and $0.50 and $2.50 above that. Anthropic says prompt caching can cut costs by up to 90% and batch processing by 50%.
One billing change deserves a flag. Anthropic said that since September 24 it has resumed charging for certain requests that its safeguards block before Claude responds, in areas where its false-positive rate is below 0.1%. Most developers will never see it. Security researchers who probe borderline code might.
Codex Pricing: Free, Plus and Three Pro Tiers
OpenAI sells Codex as a feature of ChatGPT rather than a standalone product. Its help center says Codex is included, with varying limits, in Free, Go, Plus, Pro, Business, Enterprise, Healthcare and Education plans. Free and Go get limited usage, and Codex Cloud, the remote execution environment, is not part of those two tiers. Plus, all Pro tiers, Business, Enterprise, Healthcare and Education accounts can use it.
Signing in with ChatGPT bills against your plan. Using a personal API key bills at API prices instead. That fork is the first decision every Codex user makes, and it explains why two developers with the same tool can report wildly different monthly costs.
The Pro ladder
The individual ladder now has three Pro rungs:
- Pro $100: introduced October 9, 2026, alongside revised Codex usage rules for Plus and Pro.
- Pro $200: still current at $200 per month. Customers who held it in the September 22 to 29 window keep their earlier allowance through October 29.
- Pro $500: introduced by October 16, 2026, with an allowance OpenAI describes as 25 times the ChatGPT Plus allowance, plus Ultrafast access.
OpenAI does not publish a single Codex-only subscription price. Its help pages describe ChatGPT Work and Codex usage as token-priced, with a credit-based rate card that was updated on October 24. According to that rate card, GPT-6.1 Sol runs at 50 credits per million input tokens and 250 per million output tokens, while the older GPT-5.3-Codex sits at 43.75 and 350. The rate card does not translate credits into dollars in the text we captured, so we do not convert them here.
Models and naming
The model lineup is the messiest part of the Codex story. One OpenAI help page lists GPT-6.1 Sol, GPT-6 Sol and GPT-6 Luna for Work and Codex. Another lists GPT-6 Astra for Pro, Business and Enterprise, and says Codex uses GPT-5.6 Terra, Sol and Luna depending on plan. OpenAI also removed GPT-5.4 and GPT-5.4 mini from ChatGPT-authenticated Codex on August 31 and replaced them with GPT-5.6 Terra and Luna. Check your own plan’s model picker before you promise anyone a specific model.
Teams already weighing Codex against editor-based agents will find the trade-offs laid out in our comparison of Codex, Cursor and Claude Code. OpenAI’s own developer news feed tracks the plan changes as they land.
Gemini CLI Pricing: Free Software, Paid Model Access
Gemini CLI is the odd one out. The project describes itself as fully open source under the Apache 2.0 license, and Google’s Gemini CLI project page lists the repository and release channels. There is no Gemini CLI subscription to buy. What you pay for is model access.
The README text we captured still advertises Gemini 2.5 Pro with a 1 million token window and a free tier of 100 requests per day. That wording predates Gemini 4 Argon by a generation, and our sources did not confirm the current free quota. If you plan around that number, check Google’s live documentation first.
Argon API economics
Gemini 4 Argon reached developers on October 13 through the paid tier of the Gemini API. Google lists launch pricing of $2 per million input tokens and $10 per million output tokens, with cached input tokens 95% cheaper than the normal input rate. The introductory period ends November 30. From December 1 the stated price is $4 and $20, and Google says that applies to the broadly available Argon model. Google also raised the output ceiling from 64,000 tokens to 1 million.
Notice the December math. At $4 and $20, Argon’s list price matches the standard Opus 5.5 rate that Anthropic reportedly publishes. The price advantage Gemini has this month disappears at the start of next month, and the real contest moves to cache discounts and task efficiency.
Subscriptions and Antigravity
Google also sells Google AI Plus, Pro and Ultra subscriptions. Its plan pages describe Pro with 5 TB of storage, higher Gemini usage limits and $40 in monthly Google Cloud credits, at a displayed U.S. price of $19.99 per month. Ultra tiers list 20 TB or 30 TB of storage and up to 20 times the Pro usage limits. Those figures come from a single source set, so confirm them on Google’s pages before buying.
Google says Argon is available inside its Antigravity agent platform for Ultra subscribers, and an October 23 update reportedly extends that to Pro. We could not confirm whether a given subscription tier raises Gemini CLI quotas. For the open-source side of the field, our guide to OpenCode, Codex CLI and Gemini CLI covers how community tooling changes the cost picture.
Price Tables: Plans and Token Rates Side by Side
Two tables follow. The first covers subscription and plan prices that vendors publish. The second covers per-token rates. Where a vendor has not published a price in the sources we verified, the cell says so.
| Plan or tier | Vendor | Price | Notes |
|---|---|---|---|
| Claude Pro | Anthropic | $20 monthly, or $200 billed annually up front | First paid individual plan |
| Claude team seat | Anthropic | $20 per seat per month annual, $25 monthly | Claude Code on every seat, up to $500 per month pooled API credits |
| Claude higher seat | Anthropic | $125 per seat per month on monthly billing | Weekly limits apply on premium seats |
| ChatGPT Free and Go | OpenAI | Free and Go pricing not confirmed here | Limited Codex, no Codex Cloud |
| ChatGPT Pro 100 | OpenAI | $100 per month | Introduced October 9, 2026 |
| ChatGPT Pro 200 | OpenAI | $200 per month | Allowance drops after October 29 for grandfathered users |
| ChatGPT Pro 500 | OpenAI | $500 per month | 25 times Plus allowance, Ultrafast access |
| Gemini CLI | $0 for the software | Apache 2.0, you pay for model access | |
| Gemini API paid tier | Per token, see next table | Argon available October 13 |
| Model | Vendor | Input per 1M tokens | Output per 1M tokens |
|---|---|---|---|
| Claude Opus 5.5 Fast mode | Anthropic | $8 | $40 |
| Claude Haiku 5.5 (prompts up to 100,000 tokens) | Anthropic | $0.10 | $0.50 |
| Claude Haiku 5.5 (prompts above 100,000 tokens) | Anthropic | $0.50 | $2.50 |
| Gemini 4 Argon (through November 30) | $2 | $10 | |
| Gemini 4 Argon (from December 1) | $4 | $20 |
Anthropic’s standard Opus 5.5 rate of $4 and $20 rests on one source, so it appears in the text above and not in this table. OpenAI’s Codex credits have no dollar conversion in the pages we fetched, which is why no OpenAI row appears in the token table. For consumer chatbot plans outside coding agents, our roundup of chatbot subscription prices lines up the monthly tiers.
Speed Premiums: Fast Mode, Ultrafast and What They Cost
Speed has become a priced feature. Both Anthropic and OpenAI now sell a faster tier, and the discounts that matter most apply to the slower default.
Anthropic describes Fast mode as a high-speed configuration for Opus 5.5 that delivers lower latency at a higher cost per token. The company says it runs up to 2.5 times faster at twice standard pricing, and it is in research preview inside Claude Code. You pay twice per token for at most 2.5 times the speed, so the premium buys time and not tokens. For interactive work where a developer waits on every response, the premium can pay for itself. For overnight batch refactors it cannot.
OpenAI’s Ultrafast is more extreme. The company says it produces tokens up to eight times faster in Codex. It is available at launch on Pro 500 and in eligible Enterprise and Education workspaces, and it is not available on Plus, Pro 100, Pro 200 or Business, even through credits. On Pro 500, Ultrafast first consumes included usage at eight times the standard rate, then draws credits at six times the standard credit rate. A developer who leaves Ultrafast on all day will burn through the allowance far faster than the plan headline suggests.
Google has not announced a CLI speed tier. What it offers instead is cheap cached input. A 95% discount on cached tokens rewards long sessions where the agent re-reads the same files, and that pattern describes most agent work. The same logic applies to Anthropic, where cache reads cost a small fraction of fresh input. Our comparison of API cache pricing across Gemini, ChatGPT and Claude goes deeper on that point.
Benchmarks and Vendor-Reported Numbers
Here is the uncomfortable part. We could not find a neutral, dated head-to-head benchmark that runs Claude Code, Codex and Gemini CLI on the same tasks with October 2026 models. Earlier leaderboard numbers in circulation use older model generations or vendor harnesses, so we left them out. What exists are vendor-reported figures from three sources, and they measure different things.
| Source | Reported figure | What it measures | Caveat |
|---|---|---|---|
| Google (Gemini 4 Argon launch) | 77.9% on DeepSWE v1.1 | Real-world long-horizon software engineering tasks | Vendor-run, no matching Claude or Codex score published |
| Google (Gemini 4 Argon launch) | 68% on CWE-bench v1, tied for first | Fixing security vulnerabilities | Tie means other models match it |
| Anthropic (Opus 5.5 and Sonnet 5.5 pages) | 40% lower cost than Opus 5, Sonnet 5.5 30%+ faster | Cost and speed versus the vendor’s own previous models | Compares Anthropic with Anthropic |
| OpenAI (Codex customer page) | 30 to 50% shorter early iteration time at Harvey | One customer’s workflow | Single customer claim, not a benchmark |
| OpenAI (Ultrafast docs) | Up to 8x faster token generation | Raw generation speed | Pro 500 and eligible workspaces only |
Read the table as marketing context, not a ranking. Google’s DeepSWE score is the only task-completion number from the three vendors, and Google ran it. Anthropic’s numbers compare Opus 5.5 against Opus 5, which tells you the generational gain but nothing about Codex. OpenAI’s Harvey line is a customer outcome rather than a test.
For buyers, the practical benchmark is your own repository. Run the same ten tickets through each agent, record the tokens or plan usage consumed, and count how many diffs merge without edits. That cost-per-merged-change figure beats any leaderboard because it includes the price of your own review time.
Worked Cost Example: One Task at Five Price Points
Abstract rates mislead, so here is one worked example. The numbers are illustrative arithmetic on published prices, not measurements of real sessions. For a wider view of model pricing, see our Claude, Gemini and ChatGPT price comparison. Imagine an agent task that reads 1 million input tokens and writes 100,000 output tokens, which is plausible for a multi-file refactor with repeated context.
| Model and mode | Input cost | Output cost | Task total |
|---|---|---|---|
| Claude Opus 5.5 standard ($4 and $20) | $4.00 | $2.00 | $6.00 |
| Claude Opus 5.5 Fast mode ($8 and $40) | $8.00 | $4.00 | $12.00 |
| Gemini 4 Argon through November 30 ($2 and $10) | $2.00 | $1.00 | $3.00 |
| Gemini 4 Argon from December 1 ($4 and $20) | $4.00 | $2.00 | $6.00 |
| Claude Haiku 5.5, prompts above 100,000 tokens ($0.50 and $2.50) | $0.50 | $0.25 | $0.75 |
Prices ignore caching, which would cut the input line sharply on all of these. They also ignore the fact that a cheaper model may need more retries. A $0.75 Haiku task that fails twice costs more than a $3 Argon task that lands first time.
Here is a small script that reproduces the table and lets you swap in your own token counts:
RATES = {
"opus-5.5": (4.00, 20.00),
"opus-5.5-fast": (8.00, 40.00),
"argon-intro": (2.00, 10.00),
"argon-dec": (4.00, 20.00),
"haiku-5.5-long": (0.50, 2.50),
}
def task_cost(model, input_tokens, output_tokens):
inp, out = RATES[model]
return input_tokens / 1_000_000 * inp + output_tokens / 1_000_000 * out
for name in RATES:
print(name, round(task_cost(name, 1_000_000, 100_000), 2))Now scale it. If a developer ran many such tasks a month, per-token API metering would add up quickly. Against that, a $20 or $100 or $200 subscription looks cheap, which is exactly why flat plans exist and why vendors meter them with weekly limits.
Real-World Examples: Who Is Paying for What
Pricing pages describe lists. These teams show how the products get bought in practice, using announcements from the past month.
- Barclays and Claude Code. On October 1, Anthropic said Barclays expected Claude Code adoption to reach 50% of its developer population by the end of 2026, with a majority of software engineers by 2027. A bank on that curve buys seats, not tokens.
- Cognizant’s Flowsource platform. Anthropic expanded its Cognizant partnership to run Claude Code inside Flowsource, Cognizant’s software-engineering platform. The announcement was updated on October 21.
- Harvey and Codex. OpenAI’s Codex page quotes a Harvey line: “At Harvey, Codex transformed how we build by cutting early iteration time by 30â50%, freeing engineers to focus on system design and high-leverage decisions.” That is a customer outcome OpenAI chose to publish, so weigh it as marketing.
- Google’s own engineers on Argon. Google says its engineers use Argon for everyday debugging, large-scale codebase migrations and algorithm design. A vendor dogfooding its model is a signal about confidence, not about price.
- Genesis Mission research projects. On October 8, Anthropic committed $150 million over three years to the Genesis Mission and said it would provide Claude, Claude Code and API credits to several hundred research projects. It also opened 10,000 free and discounted Claude seats for academic scientists.
- Anthropic’s own source-control rebuild. Anthropic published an account of redesigning its internal version control and said it uses Claude Code extensively for internal software development.
- Codex on Google Cloud. Google Cloud said on October 22 that some enterprise customers could join a private preview of OpenAI models in Model Garden and connect local Codex CLI workflows to Gemini platform inference. A Codex front end running on Google’s models is a pricing wildcard worth watching.
The pattern across these examples is that large organizations negotiate seats and credits, while individual developers choose between plan tiers and API keys. The mid-size team that mixes both is where billing surprises happen.
Use-Case Recommendations: 6 Developer Profiles
Match the plan to the work pattern, not to the brand.
The weekend builder. Start with Codex on a Free or Go account to test the workflow at no cost, then move to a $20 Claude Pro plan if you hit limits. Both let you learn the tools before paying real money.
The full-time solo developer. A Claude subscription or ChatGPT Pro $100 tier is the sweet spot. Pro $100 is new this month and sits between Plus and the $200 tier, which fills a gap that pushed many users toward API keys.
The heavy parallel-agent user. If you run several agents at once, price Pro $500 or a Claude premium tier against API tokens using the worked example above. The break-even sits where your monthly API estimate crosses the plan price. For editor-centric alternatives in this tier, see our look at Kiro versus Claude Code.
The platform team that audits its tools. Gemini CLI’s Apache 2.0 license lets you read the code, fork it and pin versions. The project promotes a stable release each week on Tuesdays at 20:00 UTC, so pinning matters.
The security-focused engineer. Google reports Argon tied for first on CWE-bench v1 at 68%. That is a vendor claim, so test on your own code.
The enterprise buyer. Look at Claude Team seats with pooled credits, ChatGPT Business or Enterprise with workspace credits, or Google Cloud billing for Gemini. The deciding factor is usually where your cloud commitment already sits.
Pros and Cons of Each Agent
Claude Code
- Pros: clear individual ladder from $20, per-seat team pricing with pooled credits, Opus 5.5 that Anthropic says costs 40% less to run than Opus 5 on typical workloads, published Haiku price for cheap subagents, usage limits raised this month.
- Cons: Fast mode doubles the price for at most 2.5x speed, weekly limits still apply on heavy use, no free Claude Code tier confirmed, billing now applies to some safeguard-blocked requests.
OpenAI Codex
- Pros: included on every ChatGPT plan including Free and Go, Codex Cloud with no separate VM charge at launch, three Pro rungs from $100 to $500, Ultrafast for latency-sensitive work.
- Cons: no standalone price, credits with no dollar conversion in the fetched pages, conflicting model lists across help pages, Ultrafast limited to the top tier.
Gemini CLI
- Pros: Apache 2.0 license, lowest published token price before December, 95% cached-input discount, 1 million token output ceiling on Argon.
- Cons: price doubles on December 1, no confirmed CLI quota tiers, README text looks stale, Argon requires the paid API tier.
The broader field also includes editor-first products, and our software category hub tracks the rest of the developer tooling coverage.
Migration Guide: Moving Between Agents Without Losing a Week
Switching costs are mostly configuration, not code. Here is a sequence that keeps a team productive during the move.
- Export your instructions. Copy your project rules from the current agent’s instruction file, such as CLAUDE.md, into the equivalent for the target tool, such as AGENTS.md for Codex or GEMINI.md for Gemini CLI.
- Port tool connections. List every MCP server and shell permission the old agent uses and recreate them in the new one. Security settings rarely transfer automatically.
- Choose the billing path first. Decide between plan billing and API keys before the pilot. Codex with a ChatGPT login bills the plan, and an API key bills tokens.
- Pilot on one repository. Pick a repo with good tests so you can tell a bad diff from a bad idea.
- Record cost per merged change. Track plan usage or token counts for ten tickets on each agent and compare the merged results.
- Keep the old seat for a month. Overlap the subscriptions so a missing feature does not stall a release.
- Set a December reminder. If you chose Argon for its introductory rate, schedule a review before the November 30 deadline.
Teams moving from editor assistants to terminal agents should expect the first week to feel slower, because the review habit changes from line-by-line accepts to whole-diff reads.
Verdict: Which Pricing Model Wins
No single agent wins on price for every developer. The data points to a split decision.
For individuals, Claude Code has the cleanest ladder, starting at $20 monthly and moving through higher tiers with published limits. Codex has the cheapest way in, because limited access sits on Free and Go, and the widest top end, because Pro 500 reaches $500. Gemini CLI has the lowest token price today, but a $2 and $10 rate that doubles in five weeks is a promotion, not a plan.
For teams, Claude’s $20 annual seat with pooled credits is the most transparent team structure in the sources we verified. OpenAI’s credit rate card is flexible but harder to forecast without a dollar conversion. Google’s model suits companies already invested in Google Cloud billing.
Our recommendation: start every evaluation with the worked example, run ten of your own tickets, and pick on cost per merged change. If you want a three-way view that includes GitHub’s assistant, our comparison of Claude Code, Codex and GitHub Copilot adds the fourth option many teams already pay for.
FAQ
How much does Claude Code cost per month?
Anthropic’s pricing page lists the first paid individual plan at $20 per month on monthly billing, or $200 billed up front annually. Higher Max tiers sit above it, and team seats start at $20 per seat per month on annual billing. Heavy users can also bill API tokens through a Claude Console account.
Is Codex free?
OpenAI says Codex is included in ChatGPT Free and Go with limited usage. Codex Cloud, the remote execution environment, is not part of those plans. Plus and Pro tiers raise limits, and Pro now comes in $100, $200 and $500 versions.
Is Gemini CLI free to use?
The software is free and open source under Apache 2.0. Model usage is billed separately, and Gemini 4 Argon requires the paid tier of the Gemini API. An older README mention of a 100-request daily free tier could not be confirmed as current.
When does Gemini 4 Argon pricing go up?
Google’s introductory rate of $2 per million input tokens and $10 per million output tokens ends November 30, 2026. From December 1 the stated price is $4 and $20.
What is the cheapest Claude model in Claude Code?
Claude Haiku 5.5, announced October 7, is available in Claude Code. Anthropic prices it at $0.10 input and $0.50 output per million tokens for prompts up to 100,000 tokens, and $0.50 and $2.50 above that.
What are Claude Code Fast mode and Codex Ultrafast?
Both are paid speed tiers. Anthropic’s Fast mode for Opus 5.5 is in research preview at $8 and $40 per million tokens, up to 2.5 times faster. OpenAI says Ultrafast generates tokens up to eight times faster and is available on Pro 500 and eligible Enterprise and Education workspaces.
Which agent is best for a small team?
Claude Code’s per-seat pricing with pooled credits is the easiest to forecast. Teams already on ChatGPT Business may find Codex cheaper to add. Run a ten-ticket pilot on each and compare cost per merged change before committing.
Where can I track pricing changes as they happen?
Vendor help pages change often, and independent trackers such as Rate Limit Now log limit and pricing updates across AI coding tools. Recheck the official pages before you sign a team contract.
