Corthex · Budget / explicit assumptions
A chatbot budget you can explain.
Separate the platform, AI usage and the work your team puts in. Change the assumptions below; the estimate should change with them.
Model choices and the initial model match the current assistant selector. All replies use the same token assumption; existing prepaid balance is zero. This estimates one month, using current plan allowances and whole packs. It is not a measured workload or invoice.
Estimated AI usage
- Credits per AI reply
- 1
- Monthly subscription
- €0.00
- New prepaid purchase
- €60.00
- Purchased credits left
- 3,000
- Your maintenance budget
- €100.00
- Scenario outlay
- €160.00
Understand the unit
A conversation is not one credit.
Count AI replies and their context
A conversation can contain several assistant replies. Each reply may use the visitor’s question, earlier messages, retrieved knowledge and tool context as input. Its generated answer is output. A long conversation with extensive source material can therefore use more credits than a short exchange, even if both are counted as one conversation in an activity report. Start with explicit reply and token assumptions, then replace them with observations from your own controlled pilot.
Model choice changes the same workload
The calculator uses current Corthex model usage rates and the minimum credits for that model. Input and output token weights differ. It also follows the current long-context multiplier where the selected model has one. The result is a scenario, not a guarantee of the exact charge for a real turn. Actual consumption is reconciled from provider-reported usage; additional turns or different retrieved context change the workload.
Keep maintenance visible
Uploading a source is not the end of the work. Someone still reviews answers, updates policies, checks product exceptions and follows up handoffs. Enter your own maintenance hours and hourly budget rather than treating that work as free. This is your planning assumption, not an industry benchmark or a claim that the assistant saves a particular amount of staff time. Installation or custom integration work should be budgeted separately if your project needs it.
Current Corthex terms · reviewed 30 October 2026
A subscription and a credit purchase are different commitments.
| Route | Platform price | AI usage | What to check |
|---|---|---|---|
| Free | €0 | 25 one-time credits | A trial grant, not a monthly allowance. Limited knowledge and document capacity. |
| Pay as you go | No recurring subscription | Prepaid purchases | Purchased credits do not expire. Check assistant, document and seat limits. |
| Pro | €39 / month | 5,000 / month | Check monthly vs annual billing, included usage and the features your team needs. |
| Business | €99 / month | 20,000 / month | Check monthly vs annual billing, included usage and the features your team needs. |
Monthly billing is used above. Annual billing and higher plans are shown on the current pricing page. Human handoff is available on Pro and higher plans and requires configuration and staff availability.
Work one example all the way through.
Suppose you plan 500 conversations in one month, four AI replies per conversation and 1,500 input plus 300 output tokens per reply with GPT-6 Luna, the current new-assistant default. Under the current usage rule that scenario is 1 credit per reply, or 2,000 credits. Those are declared assumptions, not measured usage or a typical customer workload. Re-run the same calculation with longer context before deciding your budget is sufficient.
With prepaid only and a selected 5,000-credit pack, the calculator buys one whole pack to cover 2,000 credits and shows the remaining purchased balance. It does not silently assume fractional packs or automatic purchases. A different pack selection can change the upfront purchase. The tool is a comparison of the pack you choose, not a claim to optimize every possible mix of packs.
Transparent scenario
500 × 4 × 1 = 2,000conversations × AI replies × credits per reply
Then subtract the selected plan’s included credits, purchase whole packs for the remainder and add the separately entered maintenance budget.
No current wallet balance, discounts, refunds or customer-specific agreement is assumed. Check checkout totals and applicable tax treatment before purchasing.
Measure the pilot before expanding it.
Use a small question set that includes short answers, long answers, a tool-assisted question and a missing-source case. Record the model, context and actual credit consumption in your own workspace, without publishing customer conversations. Review whether the answer is useful as well as what it costs. A cheap but unsupported answer is not an acceptable substitution for a correct one, and more conversations do not by themselves establish revenue or resolved requests.
Keep the budget review tied to your launch review. If the pilot needs human handoff, multiple assistants or more document capacity, include that requirement before comparing plans. Update the scenario when sources, models or conversation length change. Downloading the assumptions makes that later comparison auditable without pretending the forecast was a measured result.