Claude vs ChatGPT vs Gemini for WordPress Development: The Ultimate 2026 Guide
Updated 21 November 2026. This guide replaces the April 2026 version, which was written around Claude Opus 4.7, GPT-5.4 and Gemini 2.5. All three vendors have shipped new model lines since, so every model name, price and limit below was re-read on the vendor’s own page today. The URL still says 2025 because it has been the same page since our first comparison.
For WordPress plugin work in November 2026, Claude is still the plan our team reaches for first, mainly because Claude Code runs in the terminal against a real repository and Sonnet-class models follow WordPress conventions well. ChatGPT Plus is a strong second tool with Codex included, and Gemini is the cheap, long-context option on the API side. The honest answer is a routing decision, not a loyalty one: a cheap model for small jobs, a workhorse for the middle, and a flagship for the hard calls. Below is what each plan and model line costs today, where each fits plugin development, and how to test them on your own code.
This post is about paid plans for people who write and maintain plugins. If you are still choosing between free chat tools, read our guide to the free plans. If you want free API access for scripts, see our free API credits guide. If you run a team and want to keep client context organised, see our Projects comparison.
Claude vs ChatGPT vs Gemini for WordPress development: which is best?
Our view after running all three on real plugin work for a long time: pick Claude as the primary, add ChatGPT Plus if you want a second opinion and a different agent, and use Gemini through the API when a huge context or a low price matters more than WordPress polish. We are not going to publish a head-to-head score table this time, because the models in our April tests no longer exist on the vendors’ current lists, and we do not want to present old results as current. What we can give you is the verified facts, our routing habits, and a test you can run in an afternoon.
| Question | Claude | ChatGPT | Gemini |
|---|---|---|---|
| Terminal coding agent | Claude Code, included from Pro | Codex (web, CLI, IDE extension), included from Plus | Antigravity, entry-level limits listed on the Pro plan |
| Entry paid plan (US, before tax) | Pro, $20 monthly or $17 a month billed annually | Plus, $20 a month | Prices not confirmed for the US today |
| Heavy-use plan | Max, $100 a month, “5x Pro usage” | Pro, $100, $200 or $500 a month | AI Ultra, with the highest Antigravity limits |
| Team plan | Team, from $20 per seat a month billed annually | Business, $20 per user a month billed annually | Google Workspace and Gemini Enterprise, per-seat and priced separately |
| Flagship API model | Claude Fable 5.2, $8 / $40 per million tokens | GPT-6.2 Astra, $8 / $40 | Gemini 4 Carbon, $2 / $10 through 30 November 2026, then $4 / $20 |
| Workhorse API model | Claude Sonnet 5.5, $2 / $10 | GPT-6.2 Sol, $2 / $10 | Gemini 4 Neon, $3 / $15 |
| Cheap API model | Claude Haiku 5.5, from $0.10 / $0.50 | GPT-6 Luna, $0.10 / $0.50 | Gemini 3.5 Flash-Lite, $0.30 / $2.50 |
Sources, read 21 November 2026: Claude pricing, Claude models overview, ChatGPT and Codex pricing, ChatGPT Pro tiers, OpenAI API models, Gemini plans and Google AI Studio pricing. Prices are per million input and output tokens unless stated.
What does each paid plan cost in November 2026?
Claude plans
Anthropic’s pricing page lists prices in US dollars and says they exclude tax. Pro is $17 a month billed annually ($200 up front) or $20 billed monthly, with more usage than Free over the same five-hour and weekly windows, access to Opus, Sonnet, Haiku and Fable (Fable is paid with usage credits on Pro), and Claude Code. Max is $100 a month with “5x Pro usage”, Fable included up to half of your weekly limit, and $100 a month in API credits. Team is for two to 150 seats: a standard seat at $20 a month billed annually ($25 monthly) and a premium seat at $100 ($125 monthly) with five times the usage. Enterprise is $20 per seat a month billed annually with a 20-seat minimum, plus usage at API rates. All three team-level plans list Projects with sharing and collaboration.
ChatGPT plans
OpenAI’s Codex pricing page lists Free at $0, Go at $8, Plus at $20, Pro “from $100” and Business at $20 per user a month billed annually ($25 billed monthly, two or more users). Pro has three price points, $100, $200 and $500 a month, according to OpenAI’s help article on Pro tiers. Codex is part of every plan, with Free and Go limited to lightweight tasks, Plus described as enough to “power a few focused coding sessions each week” across web, CLI, IDE extension and iOS, and Pro including everything in Plus with more room. One detail from the help article that surprised us: the Ultrafast mode is only on the $500 tier, and buying extra credits on the $100 or $200 tier does not unlock it.
OpenAI’s own pages are not consistent about version labels. The pricing page lists GPT-6.1 Sol and GPT-6 Luna under Plus and a feature table lists GPT-6.2 Astra and GPT-6.2 Sol as available on Plus, Pro and Business. We name models below by family (Astra, Sol, Terra, Luna) and by the labels on the API models page.
Gemini plans
The Gemini subscriptions page our machine reached was the India version with rupee prices, so we are not quoting any Gemini consumer price. What we can confirm from it: Google AI Plus, Pro and Ultra exist, they scale usage as multiples of the free plan, Pro lists “entry-level Antigravity limits” for Google’s coding agent, and the model names are Gemini 4 Hydrogen, Neon and Carbon. Check the page for your region before buying.
Which is the best AI for WordPress plugin development, Claude or GPT?
For the work our team does, which is mostly PHP plugin code, REST endpoints, WP-CLI, block code and release prep, Claude has been the better primary tool, and the deciding factor is the agent as much as the model. Claude Code reads files, runs commands and edits in place, and it composes with MCP servers we have built for our own systems. ChatGPT with Codex is now a real agent too, and we use it for a second opinion on tricky changes because a different model catches different mistakes. Our longer comparison of the agents themselves is in Codex vs Claude Code vs Cursor.
For model-level detail on the newest Claude releases, read what we know about Claude Opus 5.5 for plugin work and the same for Claude Fable 5.2. Both explain how we route work between models, which is the part that changes your bill.
The current Claude lineup
Anthropic’s models page lists four current models, all with a 1M-token context window and 128K maximum output:
- Claude Fable 5.2 (
claude-fable-5-2), $8 input and $40 output per million tokens, “for demanding reasoning and long-horizon agentic work”. - Claude Opus 5.5 (
claude-opus-5-5), $4 and $20, “for long-running agentic coding and knowledge work”. Anthropic says to start here for most workloads. - Claude Sonnet 5.5 (
claude-sonnet-5-5), $2 and $10, “the best combination of speed and intelligence”. - Claude Haiku 5.5 (
claude-haiku-5-5), from $0.10 and $0.50, for “high-volume, latency-sensitive tasks such as classification, extraction, and routing”.
Batch requests are 50 percent off and cache reads cost a fraction of the base input price, so the sticker rate is the ceiling, not the typical bill. The page also notes that 1M tokens is roughly 555,000 words on the current tokenizer, which Anthropic introduced with Opus 4.7. That is the same tokenizer change we wrote about in April: the same text now counts as more tokens than it did on older models, so compare plans using tokens measured on the current models, not an old estimate.
The current OpenAI lineup
OpenAI’s API models page lists four GPT-6 text models: GPT-6.2 Astra at $8 and $40 per million tokens (“flagship for the hardest complex reasoning and coding work”), GPT-6.2 Sol at $2 and $10 (“balanced everyday model for coding and scaled agents”), GPT-6.1 Terra at $1 and $5 (“fast, efficient model for visual reasoning and computer use”) and GPT-6 Luna at $0.10 and $0.50 (“most efficient model for high-volume, focused tasks”). The page also lists cyber-specific models for authorised security testing.
The current Gemini lineup
Google’s AI Studio pricing page lists Gemini 4 Carbon and Argon at $2 and $10 per million tokens through 30 November 2026, rising to $4 and $20 from 1 December. Gemini 4 Neon is $3 and $15, and Gemini 4 Hydrogen is $1.50 and $7.50. Gemini 3.8 Flash is $0.75 and $3.75 through 31 December, then $1.50 and $7.50. Gemini 3.5 Flash-Lite is $0.30 and $2.50. Gemini 2.5 Pro and 2.5 Flash are marked deprecated and route to newer models on 1 December 2026. If you built a workflow around the 2.5 line, plan the move now.
What is the best LLM for PHP and WordPress development?
No vendor publishes a WordPress-specific benchmark, and we will not invent one. What matters for WordPress is narrower than general coding skill:
- Does it follow WordPress Coding Standards without being told? Spacing, Yoda conditions, prefixed functions, text domains.
- Does it reach for the right API?
$wpdb->prepare, nonces and capability checks in REST callbacks,sanitize_*on input andesc_*on output. - Does it use current block APIs?
block.jsonmetadata,useBlockProps, dynamic block render callbacks, and no components that were renamed several releases ago. - Does it know when it is guessing? The worst failure is a confident, invented hook name.
Run the same brief on each model and score those four things. Here is the exact brief we reuse, and you can too:
Write a WordPress REST API endpoint at /wp-json/acme/v1/orders that accepts a POST
with product_id and quantity. Require the manage_woocommerce capability, validate
and sanitize both fields, save to a custom table with $wpdb->prepare, return a
WP_REST_Response, and register a schema for the route. Follow WordPress Coding Standards.
Then run a second brief for a dynamic block with an Inspector panel, and a third that gives the model a failing PHPUnit output and asks for the fix. Three briefs, three models, one afternoon, and you have data that matches your code instead of ours.
Does Claude Code work for WordPress development?
Yes, and it is where most of our plugin work happens. Claude Code runs on Pro and above, reads your repository, runs your test and lint commands, and edits files in place. What makes it useful for an agency rather than a single developer is the layer around it: hooks that block unsafe commands, a CLAUDE.md that carries coding standards, skills that package repeatable jobs, and MCP servers that connect it to your tickets and sites. We have written about each. Start with hooks versus CLAUDE.md rules, and see Claude Code as an agency team member for the workflows that stuck.
The limits to plan around are real. A coding agent uses a lot of tokens, and a plan’s five-hour and weekly windows are shared across everything you do on it. If several people or agents run in parallel, read how an agency plans around usage limits before you pick a plan.
How should a freelancer, an agency and an indie developer choose?
Freelance WordPress developer
Start with one $20 plan, Claude Pro or ChatGPT Plus, whichever agent you prefer, and use the free chat tiers for the rest. Add a second only when you hit limits most days. Our $20 subscription comparison goes into the tradeoffs.
Agency owner
Move to the API or to team seats with spend limits, and route by task: a Sonnet-class model as the default coding model, an Opus-class model for the hard calls, a Haiku or Luna-class model for background jobs and classification, and a cheaper long-context model for bulk offline work. Track spend per project. If you are at ten people or more and not tracking model spend per project, you are guessing. Our post on LLM cost guardrails has the four controls we start with.
Indie developer on a tight budget
Free tiers are viable for small builds. The free plans and free API credits change often, so use our two guides above and verify on the vendor page before you build on a limit.
How do we route work between models?
We are a WordPress agency with a large catalogue of plugins and themes in production, so most of our AI spend goes on routine maintenance rather than glamorous builds. The routing habit that has held up across model generations is simple:
- Default to the workhorse. A Sonnet-class model handles most tickets: a bug with a clear reproduction, a small feature, a test, a refactor with a defined boundary.
- Promote when the task is genuinely hard. Cross-file reasoning, a security-sensitive change, or a bug that survived one fix attempt goes to an Opus-class model.
- Demote the boring parts. Classification, extraction, changelog tidying and bulk text go to a Haiku or Luna-class model.
- Keep a second model for review. A different vendor’s model reading the diff catches things the author model repeats.
The point is the habit, not the names. The model names keep changing and the routing rule has not. When a new model arrives, we slot it into the same four jobs and run our ten-ticket test before changing a default.
How do you run the afternoon test yourself?
You need three things: a repository you know well, five to ten recent real tickets, and a scoring sheet. Do this once per quarter, or when a vendor ships a new line.
- Pick tickets that vary: one bug with a stack trace, one new REST endpoint, one block change, one security hardening task, one refactor.
- Give every model the same instructions and the same repository state. Do not coach one model more than another.
- Score each result on whether it passes your tests, whether it follows WordPress Coding Standards, whether it handled escaping and capability checks without prompting, and how many follow-up turns it needed.
- Record input and output tokens for each run so you can price it with the tables above.
- Decide the default and the promotion rule, and write both into your team notes.
This takes an afternoon and replaces a lot of opinion, including ours. We describe a ten-ticket version of this test, and how we route between Opus and Sonnet, in the Opus 5.5 post.
What do the API prices mean for a real coding workload?
Sticker prices only compare well when you hold the task constant. A simple way to do it: take one real ticket, run it on each model you are considering with the same instructions, and record input tokens, output tokens and whether the result passed your checks. Multiply by the table above. Three details change the answer:
- Caching. Anthropic lists cache reads at a fraction of the base input price, and an agent that replays the same system prompt and repository context benefits most.
- Batch. Anthropic lists 50 percent off for batch requests. Overnight jobs like generating readmes or auditing many sites suit it.
- Retries. A cheaper model that needs two passes can cost more than a stronger model that passes first time. Count passes, not just tokens.
The two flagship models, Fable 5.2 and GPT-6.2 Astra, are priced identically at $8 and $40, and Gemini’s top tier is priced well below them until the 1 December change. Prices of that kind are the first thing to re-check before a purchase, because the scheduled 1 December rise on Gemini 4 Carbon and Argon doubles their price.
Common mistakes people report
Checked 21 November 2026, paraphrased from press coverage and vendor pages.
- Hitting usage limits on Claude Code faster than expected. MacRumors reported users seeing rapid rate-limit drain, and DevClass reported Anthropic admitting users were hitting limits faster than expected. The lesson is to watch usage in your first week on any plan and to keep sessions focused.
- Buying a ChatGPT Pro tier for a feature it does not include. OpenAI’s help article says Ultrafast is only on the $500 tier, and credits on the $100 and $200 tiers do not unlock it.
- Building on a model that is about to be retired. Google’s pricing page marks Gemini 2.5 Pro and 2.5 Flash as deprecated with a 1 December 2026 reroute, so any script pinned to them changes behaviour that day.
- Comparing against an old token estimate. Anthropic’s models page says the current tokenizer fits fewer words in a given token budget than older ones did.
- Using a flagship model for autocomplete-grade work. With a ten-times gap between flagship and cheap-model prices, routing is where most of the savings come from.
Frequently asked questions
Which AI is best for WordPress plugin development in 2026?
For our team, Claude, mostly through Claude Code, because the agent runs in the terminal against a real repository and the models follow WordPress conventions well. ChatGPT with Codex is a good second tool. Test on your own code before you decide.
Is ChatGPT Plus worth it for WordPress development?
At $20 a month it includes Codex across web, CLI and IDE, and it is strong for brainstorming, documentation and second opinions. Pair it with another tool for production code if you want a second model checking the first.
Can Gemini write WordPress code?
Yes, and its API prices are low on the smaller models. Its newest top models are not on the free API tier, and the 2.5 line is being retired on 1 December 2026.
What is the cheapest model for bulk WordPress tasks?
The cheap tiers from Anthropic and OpenAI list the same $0.10 and $0.50 per million tokens for Claude Haiku 5.5 (from) and GPT-6 Luna. Gemini 3.5 Flash-Lite is $0.30 and $2.50. Cheap models suit readmes, changelogs and classification, not security-sensitive code.
Is it worth paying for more than one AI subscription?
For a professional developer, a second tool earns its place if it saves a couple of hours a month, mainly as a second opinion. For an agency, move to team seats or the API and route by task instead.
How often should I re-check these prices?
Monthly, before any purchase. This page was last checked on 21 November 2026, and the Gemini price change on 1 December shows how fast the numbers move.