CustomGPT.ai Review 2026: The Query-Credit Math Behind the 1,000-Query Plan
Published May 4, 2026 · Updated September 30, 2026 · by Pondero Reviews
The short version
CustomGPT.ai review, September 2026 refresh. The plan meters in query credits, Claude 5 Opus now costs 2 extra per message, and Standard pays steeper multipliers than Premium. What 1,000 credits really buys, and who should pick which tier.
Pros
- ✓Sitemap or document corpus to a working cited bot in well under an hour, no code
- ✓SOC 2 Type II and GDPR on every paid tier, which is what clears a procurement review
- ✓One agent drives the widget, REST API, Slack, and an MCP server from a single corpus
- ✓Enterprise model selector now carries GPT-5.6 and Claude 5, with flat-rate options in both families
- ✓Credit costs for every action and depth setting are published in the docs, so the math can be done before you buy
Cons
- ✕The pricing page says 1,000 queries; the product bills 1,000 query credits, and several features spend more than one per answer
- ✕Standard pays steeper action multipliers than Premium: a verified answer costs 10 credits on Standard versus 2 on Premium
- ✕The extra-queries add-on costs more per credit than the Standard plan itself
- ✕Model selector, SSO, and a signed DPA remain Enterprise-only, so the new Claude 5 and GPT-5.6 models are out of reach on self-serve tiers
- ✕No self-host: corpus and inference live on the vendor cloud
CustomGPT.ai Review 2026: The Query-Credit Math Behind the 1,000-Query Plan
By Pondero Reviews Updated 2026-09-30 against the CustomGPT.ai changelog (entry covering September 23 to 29, 2026), the vendor docs on query credits, and the live pricing pages for CustomGPT.ai, Chatbase, and SiteGPT.
Our July review of CustomGPT.ai told readers the plan "still meters usage in queries per month, not credits." That was wrong, and it was wrong in a way that costs money. The product bills in query credits, some features spend several credits per answer, and on the $99 Standard plan those multipliers are steeper than on Premium, per the vendor's actions-cost doc and pricing page. The newest changelog entry adds GPT-5.6 and Claude 5 models with their own surcharges, including Claude 5 Opus at 2 extra query credits per message, per the September 23 to 29 changelog.
In practice, a Standard buyer who switches on Verify Responses for every answer gets roughly 100 verified answers a month, not 1,000, per our arithmetic from the actions-cost doc.
Revised verdict: 3.8 of 5, and Premium is the real entry tier for most teams
We are moving the score from 4.0 to 3.8. Retrieval quality has not gotten worse. Standard is simply a smaller plan than its headline number implies, and the gap is documented only in the help docs, not on the pricing page.
For a mid-market support or knowledge-ops team that wants the anti-hallucination features CustomGPT.ai markets, Premium at $499/mo is where the credit math stops fighting you, per the pricing page. Standard at $99/mo still works for one quiet internal bot left at default settings. Under a roughly $50 budget, Chatbase or SiteGPT remain cheaper places to start. Enterprise buyers get the new Claude 5 and GPT-5.6 models, and should default to the flat-rate ones.
The correction: it has been credits for months
Three dated vendor sources contradict what we wrote in July.
The help doc on high-performance models, last updated 2026-04-07, already listed Claude Opus 4.5, Claude Opus 4.6, Sonnet 4.6 Reasoning, GPT-5.4, and Gemini 3.1 at 2 queries per message. In the actions-cost doc, updated 2026-06-11, the vendor says every interaction "consumes query credits" and that an action debits "1 query credit plus the credit for that particular action." And in the changelog entry for August 5 to 18, the Limits & Usage page switched its wording to "query credits" and "word credits" to match the Billing page.
So the in-app unit is a credit. The marketing page is what never caught up: the compare table on the pricing page, fetched 2026-09-30, still reads "1,000 queries" on Standard and "5,000 queries" on Premium. Anyone who shops from that page alone, as we did in July, will overestimate how many answers they are buying.
What shipped September 23 to 29
The changelog entry covering those dates adds two model families, per the changelog:
| Model | Extra query credits (per changelog) | Credits per message, our reading of "extra" | Source |
|---|---|---|---|
| GPT-5.6 Terra | none listed | 1 | changelog |
| GPT-5.6 Luna | none listed | 1 | changelog |
| Claude 5 Sonnet | none listed | 1 | changelog |
| GPT-5.6 Sol | 1 | 2 | changelog |
| Claude 5 Sonnet Reasoning | 1 | 2 | changelog |
| Claude 5 Opus | 2 | 3 | changelog |
Every earlier GPT-5.x variant and every earlier Anthropic model is now tagged legacy. For self-serve buyers, the key line is lower down: the new models are "available to everyone with model selector access." That access is Enterprise-only, per the "AI model selector" row on the pricing page and a note in the high-performance models doc. A Standard or Premium buyer cannot turn on Claude 5 Opus at all.
For Enterprise, the change is a price step at the top. In April the docs put Claude Opus 4.5 and 4.6 at 2 queries per message; Claude 5 Opus at 1 base plus 2 extra lands at 3, per our reading of the changelog against the April doc. Neither the pricing page nor the April doc names GPT-5.6 or Claude 5 yet. The Bedrock row on the pricing page still says "Claude 3, 3.5 Sonnet via AWS Bedrock."
Two smaller fixes in the same entry: a source file with a corrupted section now ingests everything except the unreadable part instead of failing outright, and task file API responses finally fill download_url for agent-generated files.
What 1,000 Standard credits actually buy
The multipliers that do reach Standard and Premium come from context depth and agent actions, and CustomGPT.ai publishes a separate, higher rate for Standard on almost every one of them, per the actions-cost doc and the context-depth cost doc.
| Setting or action | Credits per answer on Standard | Credits per answer on Premium | Answers from 1,000 Standard credits (our arithmetic) | Source |
|---|---|---|---|---|
| Plain chat message | 1 | 1 | 1,000 | actions-cost doc |
| Balanced context depth | 2 | 1 | 500 | context-depth cost doc |
| Extended context depth | 3 | 2 | about 333 | context-depth cost doc |
| Lead Capture | 3 | 2 | about 333 | actions-cost doc |
| Web Search (one search) | 4 | 3 | 250 | actions-cost doc |
| Verify Responses | 10 | 2 | 100 | actions-cost doc |
| Document Analyst | 15 | 10 | about 66 | actions-cost doc |
Action rows are the 1 base credit plus the listed action cost. Standard caps web search at one search per question; Premium allows up to three, at 2 extra credits each. The docs also warn that actions and premium models cost more at higher depths, so a Balanced-depth agent that also verifies will cost more than either row alone.
Example: a support bot on Standard with Verify Responses on for every answer spends about 10 credits per reply, so the 1,000-credit month covers about 100 replies. The same bot on Premium spends 2 per reply, and 5,000 credits cover about 2,500. Roughly 5x the plan price buys 25 times the verified answers, per our arithmetic from the actions-cost doc and the pricing page.
Verify Responses runs multiple checks per answer, per the context-depth cost doc, and it carries the widest gap between tiers. That single row is why the score moved.
One gap in the docs: Standard users "pay 1 more query credit than Premium and Enterprise users at the same depth level," yet only Balanced (2) and Extended (3) are spelled out for Standard, per the context-depth cost doc. Standard agents default to Compact, per the context-depth doc, and its Standard rate is unstated. If ten Compact messages move your counter by 20, your ceiling is 500 plain answers, not 1,000.
The add-on costs more per credit than the plan
When a Standard bot runs out, the self-serve lever is the extra-queries add-on: 2,500 more a month for $375/mo, or $338/mo paid annually, per the add-on doc and the pricing page. Per credit, that is 15 cents, against 9.9 cents a credit inside the $99 Standard plan, per our arithmetic from the pricing page.
Stacking one add-on on Standard costs $474/mo for 3,500 credits, per our arithmetic from the pricing page. Premium is $499/mo for 5,000 credits plus the cheaper action rates above, per the pricing page. If a Standard bot needs the add-on even once, upgrade instead. Enterprise plans cannot buy the add-on at all, per the add-on doc.
How it compares at the entry price
Chatbase meters in message credits and says so on its pricing page. SiteGPT goes further and prints a calculator stating that model choice can swing your real message count by 10x on the same plan, per the SiteGPT pricing page. CustomGPT.ai runs a credit system with the same shape, but its pricing page gives no hint that the count can shrink.
| Tool | Entry price | Monthly allowance | REST API on entry tier | Source |
|---|---|---|---|---|
| CustomGPT.ai Standard | $99/mo ($89/mo billed annually) | 1,000 query credits | Yes | pricing |
| Chatbase Hobby | $32/mo billed annually | 500 message credits | No; API starts at Standard, $120/mo | pricing |
| SiteGPT Starter | $39/mo billed yearly | Up to 4k messages | No; API starts at Growth, $79/mo billed yearly | pricing |
Two corrections to our July table live in that grid. Chatbase Hobby and SiteGPT Starter do not list API access; both vendors gate it to their next tier, per the Chatbase and SiteGPT pricing pages fetched 2026-09-30. And July said neither alternative markets SOC 2 Type II. SiteGPT's pricing page now shows SOC 2 Type II, GDPR, and HIPAA badges, and offers a signed DPA on request at Enterprise, per the SiteGPT pricing page.
That narrows the procurement argument for CustomGPT.ai. What still separates it is the REST API on the lowest paid tier, the connector list (Confluence, Notion, Zendesk, SharePoint on Premium, and Freshdesk as of the September 16 to 22 changelog), and citations on every answer.
For overflow, Chatbase sells auto-recharge credits at $40 per 1,000, per the Chatbase pricing page. Call it 4 cents a credit against CustomGPT.ai's 15-cent add-on, per our arithmetic from both vendors' pricing pages (Chatbase, CustomGPT.ai). Units differ between vendors, so compare on your own traffic.
Who should buy which tier
Knowledge-ops lead with one internal bot. A couple dozen staff questions a day against Confluence and Google Drive, no web search, no verification. Standard at default depth fits, and so does the 3-seat limit.
Support ops lead at a 200-person SaaS. A public help-center bot with Verify Responses on, because a wrong answer to a customer is worse than no answer. This is a Premium buyer from day one. On Standard, the verification rate alone would cap you near 100 replies a month, per the table above.
Enterprise platform team already on a custom plan. Set Claude 5 Sonnet or GPT-5.6 Terra as the default, both flat-rate per the changelog. Put Claude 5 Opus on a single low-volume agent where the reasoning depth earns 3 credits a message, and size the contract's credit volume with that multiplier in it.
Security and procurement (unchanged since July)
SOC 2 Type II and GDPR appear on every paid tier, while custom SSO, role-based access, and a data processing agreement are Enterprise-only rows, per the pricing page. If legal needs a signed DPA, your entry price is a sales call.
Getting started: three settings to check on day one
- Start the 7-day trial at CustomGPT.ai. A card is required, so set a reminder for day six.
- Build one agent from your help-center sitemap and leave context depth at the default. Send ten real questions, then open Limits & Usage and write down how many credits moved.
- Turn on each action you plan to use (Verify Responses, Web Search, Lead Capture) one at a time, send ten more questions, and read the counter again. Multiply the per-answer figure by your expected monthly volume. If the total clears 1,000, start on Premium.
Twenty minutes of this settles the tier question with your own traffic.
The final call
CustomGPT.ai earns 3.8 of 5. The cited-RAG product, the SOC 2 posture, and the API on every tier still make it a sound buy for mid-market teams, and the Enterprise model selector now carries current GPT-5.6 and Claude 5 options. It drops from 4.0 because the plan bills in credits, the pricing page does not say so, and a verified answer on Standard costs five times what it costs on Premium, per our arithmetic from the actions-cost doc.
Pick CustomGPT.ai Premium if you run a customer-facing bot and want Verify Responses on. Pick Standard only for a low-volume internal bot at default depth with actions off. Pick Chatbase or SiteGPT if your budget is under $50 a month and you can live without an API on the entry tier, per their Chatbase and SiteGPT pricing pages. And if you are on Enterprise, default to the flat-rate Claude 5 Sonnet or GPT-5.6 Terra, and keep Claude 5 Opus off your high-traffic agents.
Review history
This review is updated in place as the product changes. Earlier versions:
- September 30, 2026: The Query-Credit Math Behind the 1,000-Query Plan (current version)
- July 1, 2026: Pricing, Credits Model, and the $99 Wall Explained
- May 7, 2026: Clear Take + Alternatives
Ready to try it?
Try customgpt →