OpenAI quietly shut a door on September 19, 2026. Since August 16, 2026, personal ChatGPT accounts on Free, Go, Plus, and even the $200-a-month Pro tier can no longer create a new Custom GPT or publish one to the GPT Store, according to updated OpenAI Help Center documentation reported by GPTs Korea. More than 3 million GPTs exist in total, but only about 159,000 are public and active. That single policy change reframes a question a lot of software teams are asking this month: if you want a personalized, instructable AI agent baked into a chatbot you already pay for, which platform should you actually build on now — OpenAI’s GPT Store, Anthropic’s Claude Skills, or Google’s Gemini Gems?
These three features solve the same basic problem in three very different ways. Custom GPTs are a no-code configuration layer wrapped around ChatGPT. Claude Skills are folders of instructions, scripts, and resources that Claude loads like plugins, most at home inside Claude Code. Gemini Gems are saved conversational presets built into the Gemini app, with a free tier that already includes a 1-million-token context window. None of them is a drop-in replacement for the other two, and the gaps between them — in price, in who can build one, in whether you can even publish it publicly — have gotten a lot more consequential since OpenAI’s August freeze. This comparison walks through the specs, the pricing, real deployment examples, and a migration path for anyone stuck on the wrong side of that freeze.
What Custom GPTs, Claude Skills, and Gemini Gems Actually Are
Before comparing prices and feature counts, it helps to be precise about what each product actually is, because the three are not the same category of thing wearing different names.
A Custom GPT is a version of ChatGPT that a user configures with a system prompt, optional uploaded knowledge files, and a set of enabled tools — web browsing, image generation, Code Interpreter for running code, and Actions that call external APIs through a JSON schema. It lives inside the ChatGPT interface and can be published to the GPT Store, OpenAI’s public directory, where anyone on a qualifying plan can find and use it. Building one has never required writing code; it’s closer to filling out a detailed settings form.
Claude Skills work differently. Anthropic defines a Skill as a folder containing instructions, scripts, and resources that Claude loads dynamically to get better at a specialized task, with a SKILL.md file holding the metadata Claude reads to understand what the skill does. Skills are distributed as plain folders — the official anthropics/skills repository on GitHub is the canonical source, and it can be registered directly inside Claude Code as a plugin marketplace using a single command. That makes Skills closer to developer packages than to chatbot settings: they’re versioned, git-friendly, and designed to bundle actual scripts that call external APIs.
Gemini Gems sit somewhere in between conceptually, but lean toward the no-code end. A Gem is a saved custom configuration inside the Gemini app — instructions and preferences a user sets once, then reuses or shares — accessed from a dedicated panel at gemini.google.com where users can also browse Google’s own pre-made library, according to a 2026 review from AIToolAnalysis. Gems don’t expose a folder structure or a scripting layer the way Skills do; they’re closer to a persistent, nameable persona sitting on top of Gemini’s models.
The three also arrived on different timelines, which partly explains why they’ve matured into such different shapes. OpenAI’s GPT Store launched first, back in January 2024, giving it roughly two and a half extra years to accumulate the 3 million-plus GPTs cited by GPTs Korea before the August 2026 freeze capped further personal-account growth. Gemini Gems followed later in 2024 as a lighter-weight, no-code layer built directly into the Gemini consumer app. Claude Skills are the newest of the three in their current form — Anthropic’s official anthropics/skills repository was created in September 2025, per its own GitHub history, making it barely a year old and still the fastest-growing of the three by the star-count metrics available. That age gap shows up throughout this comparison: Custom GPTs have the deepest tooling and the largest public catalog, while Claude Skills have the newest architecture and the most developer-friendly distribution model.
Full Specifications Comparison
The table below lines up the core mechanics side by side, using the most recent September 2026 documentation and third-party reporting available for each platform.
| Spec | OpenAI Custom GPTs / GPT Store | Anthropic Claude Skills | Google Gemini Gems |
|---|---|---|---|
| Core format | No-code configuration (prompt, files, tools) inside ChatGPT | Folder of instructions, scripts, and resources with a SKILL.md file | Saved persona/instruction preset inside the Gemini app |
| Who can create one | Blocked for new creation on personal Free, Go, Plus, and Pro accounts since Aug. 16, 2026 | Any Claude user with Claude Code access; distributed via GitHub | Available on the free Gemini tier, per AIToolAnalysis |
| Public marketplace | GPT Store: ~159,000 public/active GPTs out of 3M+ created | No official paid marketplace yet; community hubs like secondsky/claude-skills list 145 skills | Google-curated “pre-made library” inside the app; no public count disclosed |
| Can call external APIs | Yes, via Actions (JSON-schema tool definitions) | Yes, skills can bundle scripts that call APIs (examples include Cloudflare D1 and Tailwind integrations) | Not documented as a user-exposed API/Actions mechanism |
| Can browse the web | Yes, built-in browsing tool | Depends on the harness (Claude Code / Claude.ai) rather than the skill itself | Implicit via Gemini’s Google-integrated search and Deep Research |
| File uploads / knowledge base | Yes, uploaded knowledge files per GPT | Yes, resources bundled in the skill folder | Yes, via Google Drive integration |
| Code execution | Yes, Code Interpreter / Advanced Data Analysis | Yes, skills can contain executable scripts | Not a core, user-facing feature |
| Version control | No native git integration | Yes, skills live as folders in git repos by default | No native git integration |
| Install/discovery method | Browse GPT Store inside ChatGPT | /plugin marketplace add anthropics/skills in Claude Code |
Browse pre-made library in the Gemini app sidebar |
| Builder monetization | Engagement-based revenue share, ~$0.03 per conversation on average | None official; Anthropic has floated a future marketplace | None documented |
| Cross-platform reuse | Locked to ChatGPT | Reported to also work with Codex CLI and, per SkillsLLM, ChatGPT itself | Locked to the Gemini app |
| Context window (free tier) | Varies by underlying model | Depends on the Claude plan powering Claude Code | Up to 1M tokens on the free tier, per AIToolAnalysis |
Pricing Breakdown: What It Costs to Build and Use Each One
Pricing is where the three platforms diverge the most sharply, and where the August 2026 GPT Store freeze changes the calculus. Historically, building a Custom GPT required at least a ChatGPT Plus subscription at $20 a month, with Team and Enterprise plans also qualifying, according to both a 2026 AIONX review and older reporting from Tech.co. Free-tier users could only use pre-built GPTs from the Store, not create their own. That’s still technically the access tier — but as of mid-August 2026, even Plus and Pro subscribers on personal accounts can no longer create new GPTs or publish to the Store at all, per the OpenAI Help Center documentation cited by GPTs Korea.
Claude Skills carry no separate price tag. Anthropic’s official repository is free to clone and use, and access to Skills inside Claude Code depends entirely on whatever Claude plan already grants Claude Code access — there is no additional “Skills tier” layered on top. The catch is that Skills currently have no official paid marketplace either; a widely cited FAQ from the community-maintained “awesome Claude Skills” list on GitHub states plainly that there’s no official marketplace for paid skills yet, though Anthropic has mentioned plans for one.
Gemini Gems are the cheapest to start with. A 2026 review from AIToolAnalysis lays out the tiers: the free Gemini plan gives full Gem creation, Google Drive integration, and a 1-million-token context window running on a Gemini Flash backbone. Google AI Plus, at $7.99 a month, adds more compute headroom and a better model mix along with Deep Research, Canvas, and Gemini Live voice. Google AI Pro, at $19.99 a month or $240 a year, sets the higher-end Pro-class model as the default for Gems and adds Workspace business integration and priority access.
| Plan | Monthly Price | What It Unlocks for Agent Building |
|---|---|---|
| ChatGPT Free / Go | $0 | Use pre-built GPTs only; cannot create or publish new ones |
| ChatGPT Plus | $20/mo | Historically the entry point for building GPTs; new creation now blocked as of Aug. 16, 2026 |
| ChatGPT Pro | $200/mo | Same new-GPT restriction applies to personal Pro accounts |
| ChatGPT Business/Team, Enterprise | From $20-25/seat | Org-level accounts; personal-account freeze does not describe these as blocked |
| Claude Free / Pro | $0 / $20/mo | Skills usable wherever Claude Code is available; no extra Skills fee |
| Claude Team (Standard) | $20-25/seat | Full Claude feature set plus SSO, admin controls, and Skills access |
| Gemini Free | $0 | Full Gem creation, Drive integration, 1M-token context |
| Google AI Plus | $7.99/mo | More compute, Deep Research, Canvas, Gemini Live for Gems |
| Google AI Pro | $19.99/mo ($240/yr) | Pro-class model default for Gems, Workspace integration |
Read across that table and the irony is hard to miss: Gemini Gems, the feature with the least developer-facing power, is also the only one of the three you can build entirely for free with no subscription at all. Claude Skills are technically free too, but only if you already have Claude Code access through a paid Claude plan. Custom GPTs, the most mature and capable of the three on paper, are now the hardest to start building from scratch — because personal-account creation is frozen entirely.
What a 10-Person Team Actually Pays
Sticker prices don’t tell the whole story once you scale past a single builder, so it’s worth running the math on a hypothetical 10-person team that wants everyone to have creation access, not just consumption access. These figures are simple multiplication against the sourced per-seat prices above, not an official vendor quote, so treat them as an illustrative floor rather than a guaranteed invoice.
| Platform | Tier Needed for 10 Builders | Estimated Monthly Cost | Notes |
|---|---|---|---|
| OpenAI Custom GPTs | Not purchasable for new creation on personal plans | N/A | Requires an organizational Business/Team or Enterprise plan instead, since personal-account creation is frozen |
| ChatGPT Business (Standard seats) | 10 seats at $20-25/seat | $200-$250/mo | Org-level path around the personal-account freeze |
| Claude Skills (via Claude Pro) | 10 individual Pro subscriptions | $200/mo | Cheapest path to 10 builders with Skills access, no seat minimum |
| Claude Team (Standard seats) | 10 seats at $20-25/seat | $200-$250/mo | Adds SSO, admin controls, and central billing on top of Skills access |
| Gemini Gems (Free tier) | 10 free Gemini accounts | $0/mo | Full Gem creation with no subscription required, per AIToolAnalysis |
| Gemini Gems (Google AI Pro) | 10 seats at $19.99/mo | $199.90/mo | Only needed if the team wants the Pro-class model as the Gems default |
The gap is stark once you strip away the marketing language. A team that wants ten people building on Gemini Gems can do it for nothing, at least on the model quality the free tier offers. The same team building on Claude Skills pays for Claude access it may well already have for other reasons, since Skills carry no additional fee. The same team trying to build on Custom GPTs today has to route around the personal-account freeze entirely by provisioning an organizational plan, even if all ten people would have happily paid for Plus individually.
The August 2026 GPT Store Freeze, Explained
This is the news hook that makes the comparison timely, so it’s worth walking through carefully. On August 16, 2026, OpenAI updated its Help Center documentation to fully block the creation of new Custom GPTs and publishing to the GPT Store for personal Free, Go, Plus, and Pro accounts, according to reporting from the Korean outlet GPTs Korea. The same reporting notes that more than 3 million GPTs had been created in total by that point, but only around 159,000 were active and public in the Store — meaning the vast majority of GPTs ever made were already private, abandoned, or link-shared rather than publicly discoverable.
OpenAI has not published a detailed public explanation of the freeze alongside the documentation change, and the sourcing available doesn’t confirm whether it’s a permanent architectural shift, a temporary pause tied to abuse or spam cleanup, or a step toward folding Custom GPT functionality into newer surfaces. What is confirmed is narrower: existing GPTs continue to function, the underlying tool set (browsing, Code Interpreter, file uploads, image generation, Actions) hasn’t been removed from the product, and the restriction is specifically on new creation and publishing from personal accounts. For builders who relied on the GPT Store’s revenue-share program — which pays out roughly $0.03 per conversation on average, according to a 2026 business guide from DigitalApplied — the freeze effectively caps the ecosystem at its current size unless they have access through an organizational account.
That’s the backdrop against which Claude Skills and Gemini Gems look more attractive than they might have a year ago. Neither requires OpenAI’s approval process, neither has a comparable public freeze, and both remain fully open to new creation as of this writing.
Adoption and Ecosystem Size: Benchmarking the Three Marketplaces
None of the three platforms publish a standardized, apples-to-apples performance benchmark the way LLM vendors publish SWE-bench or MMLU scores — there is no independent lab currently scoring “GPT quality” against “Skill quality” against “Gem quality” on a shared task suite. What does exist, across multiple independent sources, is ecosystem-size data that functions as a rough proxy for adoption and momentum.
| Metric | OpenAI GPT Store | Claude Skills | Gemini Gems |
|---|---|---|---|
| Total items ever created | 3M+ (AIONX, GPTs Korea) | 145 skills in the largest community collection, secondsky/claude-skills, v3.9.0 (Sept. 9, 2026) | Not disclosed publicly |
| Public/active listings | ~159,000 (SEO.ai analysis, cited by AIONX and GPTs Korea) | No formal public marketplace; distribution is via GitHub repos | Google-curated pre-made library, no count published |
| Community traction signal | Revenue-share payouts averaging ~$0.03/conversation (DigitalApplied) | 168,438 GitHub stars on anthropics/skills as of Aug. 12, 2026 (Moclaw.ai) | 1M-token free-tier context window, cited as a headline differentiator (AIToolAnalysis) |
| Growth trend | Frozen for new personal-account creation since Aug. 16, 2026 | anthropics/skills grew from ~132,000 stars (May 2026 snapshot, HackMD) to 168,438+ (Aug. 2026) — roughly 27% in about three months | Free tier expanded to full Gem creation with a 1M-token window by mid-2026 |
The star-count trajectory on anthropics/skills is the closest thing to a hard adoption benchmark in this comparison, and it’s a genuinely fast climb for a repository that, as Moclaw.ai points out, ships no application code of its own — it’s pure documentation and folder templates. That’s a meaningfully different growth pattern from the GPT Store, which added its 3 millionth GPT organically over roughly two and a half years before the creation freeze capped it. Gemini Gems, by contrast, doesn’t have a comparable public counter at all; Google has kept usage figures private, which makes it the hardest of the three to size from the outside.
Model Access: Which Flagship Models Power Each Feature
A Custom GPT, a Claude Skill, and a Gemini Gem are only as capable as the model underneath them, and access differs by tier. Custom GPTs generally inherit whatever model the user has selected in ChatGPT proper, meaning a Plus subscriber’s GPT can draw on more capable models than a Free-tier session running a pre-built GPT. OpenAI’s flagship API model, GPT-6 Astra, launched September 3, 2026, at $10 per million input tokens and $50 per million output tokens on the standard API tier, according to a September 2026 OpenAI pricing breakdown from CloudZero — but the available sourcing does not confirm that Astra has been wired into the consumer Custom GPT builder as a selectable backbone yet.
Claude Skills don’t carry their own model; they’re a capability layer on top of whatever Claude model is running inside Claude Code, so a Skill behaves differently on a Free account than on a paid Claude plan with access to larger context windows and heavier usage allowances. Anthropic’s September 10, 2026 update to the claude-api skill, which added a “Managed Agents auto permission policy,” signals that Anthropic is actively tuning how much autonomy Skills get to exercise without a human confirming each step.
Gemini Gems are the most explicitly tiered of the three on model access. The AIToolAnalysis review found that free-tier Gems run on a Gemini Flash-class backbone with a 1-million-token context window, while Google AI Pro subscribers get a Pro-class model as the default for their Gems — a meaningful quality jump for anything requiring deeper reasoning, not just long context.
Security and Prompt Injection Risk Across the Three Platforms
Any system that lets an AI agent read instructions from a third party — a knowledge file, a skill folder, a shared Gem — inherits some version of prompt injection risk, where malicious content embedded in a document or web page tricks the agent into ignoring its original instructions. None of the sourcing reviewed here provides a formal, apples-to-apples security audit comparing the three platforms, but each has documented characteristics worth flagging.
Custom GPTs with Actions enabled can call arbitrary external APIs based on a JSON schema the builder defines, which is powerful but also the widest attack surface of the three — a compromised or poorly scoped Action can leak data or take unintended steps on a connected service. Claude Skills carry a related risk because they bundle actual executable scripts; installing a community skill from an untrusted GitHub repository means running someone else’s code inside your Claude Code environment, the same trust model as installing any open-source package. Anthropic’s move to add a “Managed Agents auto permission policy” to its own claude-api skill in September 2026 is a direct, if narrow, acknowledgment that finer-grained permission controls are needed as Skills take on more autonomous actions.
Gemini Gems have the smallest documented attack surface of the three, largely because they don’t expose a user-configurable API-calling mechanism the way Custom GPT Actions do. That’s arguably a security feature, but it’s also why Gems can’t do things Custom GPTs and Skills can — the tradeoff between capability and exposure runs in a fairly straight line across all three products.
Common Mistakes When Building on Each Platform
Each platform has its own failure pattern that shows up repeatedly once teams move past a quick demo and into something people rely on daily.
- Custom GPTs: Treating Actions as a set-and-forget integration. Because a Custom GPT has no version control, a builder who edits the live configuration to fix one problem can silently break something else with no rollback path — test changes in a duplicate GPT before overwriting a GPT people already depend on.
- Custom GPTs: Assuming personal-account creation will come back before building a workaround. Teams still planning around eventually publishing a new personal GPT should confirm current access through an organizational plan instead, given the documented Aug. 16, 2026 restriction.
- Claude Skills: Installing a community skill without reading the scripts inside it first. Since a Skill can bundle executable code, pulling one from an unfamiliar GitHub repository carries the same risk as installing any unaudited open-source package.
- Claude Skills: Over-scoping a single skill to do too many unrelated tasks. The folder-based format works best when each skill handles one specialized job well, mirroring how Anthropic’s own document-creation skills are split by file type rather than bundled into one do-everything skill.
- Gemini Gems: Expecting Gems to call an external API the way a Custom GPT Action or a Claude Skill script can. Current documentation doesn’t describe a user-exposed Actions-equivalent for Gems, so any workflow requiring a live third-party API call needs to happen outside the Gem.
- Gemini Gems: Loading a Gem’s knowledge base with more documents than the task actually needs. The 1-million-token context window is generous, but stuffing every available document into a Gem’s context can dilute relevance compared with a tightly scoped, purpose-built Gem.
Real-World Deployment Examples
Here’s how teams are actually putting each of these three systems to work, based on the documented capabilities and use patterns each platform supports.
- Marketing agency brand-voice assistant. A content team builds a Custom GPT loaded with brand style guides as knowledge files and an Action wired to their CRM’s API, so writers can request on-brand drafts without leaving ChatGPT. This is the kind of deployment the August freeze now blocks for any team without existing organizational access, since it requires creating a new GPT.
- Engineering team document automation. A software team registers anthropics/skills as a Claude Code plugin marketplace and installs the document-creation skill bundle, letting Claude generate formatted Word, PowerPoint, and PDF reports directly from sprint data without a separate document-generation pipeline.
- Solo GPT Store builder pivoting. A consultant who previously earned in the $10-$100/month range publishing a niche SEO-audit GPT — in line with the “casual creator” band described by DigitalApplied — can no longer publish new GPTs after August 16, 2026, and is now rebuilding equivalent logic as a Claude Skill distributed through a personal GitHub repo instead.
- Support team knowledge-base Gem. An internal support desk creates a free-tier Gemini Gem connected to their Google Drive-hosted knowledge base, relying on the 1-million-token context window to ingest the full support documentation set for first-line ticket triage.
- Developer scaffolding with community skills. A small dev shop installs the Cloudflare D1 and Tailwind v4/shadcn skills from the secondsky/claude-skills community collection to speed up new project scaffolding inside Claude Code, treating Skills as reusable, version-controlled starter kits rather than one-off prompts.
- Workspace admin standardizing on Gems. An IT admin at a Google Workspace organization deliberately assigns Google AI Pro seats to the operations team specifically so their internal runbook Gems default to the higher-quality Pro model rather than the free-tier Flash backbone.
- Cross-harness skill reuse. A platform engineering team builds a single Claude Skill for internal API documentation lookups and, per SkillsLLM’s compatibility notes, reuses the same skill folder across Claude Code and Codex CLI rather than maintaining separate configurations per tool.
Who Should Use Custom GPTs
Custom GPTs still make sense for organizations that already have Team or Enterprise-level ChatGPT access, since the reported freeze specifically targets personal Free, Go, Plus, and Pro accounts rather than organizational plans. If your company already has an established GPT and simply wants to keep using or lightly tweaking it, nothing about the freeze breaks that. Custom GPTs also remain the most feature-complete no-code option of the three — browsing, image generation, Code Interpreter, file uploads, and Actions are all available in a single interface without touching a terminal, which matters for non-technical teams like marketing or customer success. For anyone starting from zero on a personal account, though, that door is currently closed.
Who Should Use Claude Skills
Claude Skills fit developer and engineering-adjacent teams best, precisely because they’re distributed as git-friendly folders rather than settings screens. If your organization already reviews code changes through pull requests, a Skill fits that workflow naturally — it’s a folder that can be diffed, versioned, and rolled back like any other piece of infrastructure. Skills are also the only one of the three that’s free to build regardless of account type, as long as you have Claude Code access, and the community ecosystem around anthropics/skills and collections like secondsky/claude-skills is growing quickly. The tradeoff is that Skills currently lack a mainstream, non-technical discovery experience comparable to the GPT Store or Gemini’s pre-made library — finding and installing one still runs through GitHub and Claude Code’s plugin commands.
Who Should Use Gemini Gems
Gemini Gems are the right fit for teams that want the lowest possible barrier to entry and don’t need external API integrations. Because full Gem creation is available on Gemini’s free tier with a 1-million-token context window, per AIToolAnalysis, Gems are an easy on-ramp for internal knowledge bases, document-heavy workflows tied to Google Drive, and simple persona-based assistants that don’t need to reach outside the Google ecosystem. Gems are also the simplest of the three to hand to non-technical staff, since there’s no folder structure, script, or Action schema to configure. The ceiling is lower, though — without a documented Actions-equivalent, Gems can’t be wired into arbitrary third-party APIs the way Custom GPTs and Skills can.
Migration Guide: Moving From a Custom GPT to Claude Skills or Gemini Gems
For anyone locked out of creating new GPTs by the August 2026 freeze, migrating the logic of an existing GPT to Claude Skills or Gemini Gems is a manageable project. Here’s the practical path.
- Export your Custom GPT’s system prompt/instructions in full — copy the exact configuration text from the GPT’s builder settings before you lose access to edit it further.
- Inventory every knowledge file attached to the GPT and download a local copy of each one.
- List every Action your GPT calls, including the API endpoint, authentication method, and JSON schema for each one.
- Decide your target: pick Claude Skills if you need code execution or API calls preserved, or Gemini Gems if the GPT was mostly a persona/instruction layer without external Actions.
- For a Claude Skills migration: create a new folder, add a
SKILL.mdfile, and paste your original system prompt into the instructions section, adapting formatting to Anthropic’s documented structure. - Recreate any Actions as scripts inside the Skill folder, translating each API call into a script Claude can execute rather than a schema-based tool definition.
- Register your new skill folder locally by copying it into
.claude/skills, or publish it by registering a personal GitHub repo as a Claude Code plugin marketplace using/plugin marketplace add <your-repo>. - Test the skill inside Claude Code with representative prompts drawn from your GPT’s original conversation logs, if you still have access to them.
- For a Gemini Gems migration: open the Gems panel at gemini.google.com, create a new Gem, and paste your adapted instructions directly into the Gem’s configuration.
- Re-upload your knowledge files through Gemini’s Drive integration rather than a direct file-attachment mechanism, since Gems lean on Drive for document access.
- Drop any Actions-based functionality that required a direct external API call, since Gems don’t expose an equivalent mechanism in current documentation — plan to handle those integrations outside the Gem itself.
- Share the finished Gem with your team through Gemini’s built-in sharing options and gather feedback before retiring the original GPT.
The honest caveat here is that neither destination is a perfect like-for-like replacement. Claude Skills preserve the most functional depth (code execution, external calls) but demand more technical comfort. Gemini Gems are easier to hand off to non-technical teammates but will force you to drop any Actions-based integrations entirely.
Pros and Cons of Each Platform
OpenAI Custom GPTs / GPT Store
- Pro: Most feature-complete no-code builder — browsing, image generation, code execution, and Actions in one interface
- Pro: Established public marketplace with 159,000 active listings for discovery
- Pro: Only one of the three with a live, if modest, revenue-share program for builders
- Con: New creation and publishing blocked for all personal accounts since Aug. 16, 2026
- Con: Revenue share pays a thin average of roughly $0.03 per conversation
- Con: No version control; changes overwrite the live GPT directly
Anthropic Claude Skills
- Pro: Completely free to build and use beyond whatever Claude plan you already have
- Pro: Git-native, versioned, and auditable — fits existing developer workflows
- Pro: Fast-growing community ecosystem (168,438+ GitHub stars on the official repo as of Aug. 12, 2026)
- Con: No official paid marketplace or monetization path yet
- Con: Discovery runs through GitHub and Claude Code plugin commands, not a mainstream app store
- Con: Installing community skills means trusting someone else’s executable code
Google Gemini Gems
- Pro: Full creation available for free, including a 1-million-token context window
- Pro: Simplest, most non-technical interface of the three
- Pro: Deep native integration with Google Drive and Workspace documents
- Con: No documented Actions-equivalent for calling arbitrary external APIs
- Con: No public marketplace or count of available pre-made Gems
- Con: No monetization path for builders
The Verdict: Which One Should You Build On?
There isn’t a single winner here, because the three platforms aren’t actually competing for the same job anymore — the August 2026 freeze has pushed them further apart rather than closer together. If your organization already has Team or Enterprise-level ChatGPT access, Custom GPTs remain the most capable no-code option and the only one with a functioning, if thin, revenue-share program at roughly $0.03 per conversation. If you’re a developer or engineering team that wants version control, code execution, and free access tied to a plan you already pay for, Claude Skills — backed by an official repository that grew from about 132,000 to more than 168,000 GitHub stars in roughly three months — is the fastest-moving option and the only one of the three still fully open to new personal-account creation without restriction. If you want the lowest possible barrier to entry, a free 1-million-token context window, and don’t need to call external APIs, Gemini Gems get a working assistant in front of non-technical users the fastest.
The one data point that should shape everyone’s near-term decision is the freeze itself: 159,000 active public GPTs is a real ecosystem, but it’s now a capped one for anyone building from a personal account. Claude Skills and Gemini Gems are the two doors still open.
Frequently Asked Questions
Can I still use Custom GPTs I created before August 16, 2026?
Yes. The documented restriction applies to creating new GPTs and publishing to the Store from personal accounts; existing GPTs continue to function.
Does the GPT Store freeze affect ChatGPT Team and Enterprise accounts?
The reporting on this freeze specifically names personal Free, Go, Plus, and Pro accounts. It does not describe organizational Team or Enterprise accounts as affected.
Do I need to know how to code to build a Claude Skill?
Not strictly — a Skill can be just instructions and resources in a folder with a SKILL.md file. But Skills that bundle scripts to call external APIs or run code do require at least basic scripting familiarity to build or safely modify.
Can I make money from a Gemini Gem or a Claude Skill?
Not currently through an official program. OpenAI’s GPT Store is the only one of the three with a live revenue-share model. Anthropic has mentioned future plans for a Skills marketplace but has not launched one.
Is a free Gemini account enough to build a usable Gem?
According to AIToolAnalysis’s 2026 review, yes — the free tier includes full Gem creation, a 1-million-token context window, and Google Drive integration, without requiring a Google AI Plus or Pro subscription.
Can a Claude Skill call an external API like a Custom GPT Action can?
Yes, in practice — Skills can bundle scripts that make external API calls, as seen in community skills that integrate with services like Cloudflare D1. The mechanism is different (executable scripts versus a JSON-schema Action definition) but the outcome is comparable.
Where do I find pre-built Claude Skills to install?
The official starting point is Anthropic’s anthropics/skills repository on GitHub, which can be registered directly as a Claude Code plugin marketplace. Community collections like secondsky/claude-skills and the “awesome Claude Skills” list add further options.
What happens to my GPT Store revenue if I can no longer publish new GPTs?
Revenue tied to existing, already-published GPTs is not reported as affected by the freeze. The restriction blocks new GPTs from being created or published, which caps future growth of a personal builder’s portfolio rather than cutting off earnings from GPTs already live in the Store.
