04Pricing
No per-seat charge. A service fee, compute at published rates, and a cap you set.
Priced as a managed service, not as software seats. You set the cap; we optimise the cost on our side. Compute is itemised at published model rates so you see exactly what the infrastructure cost.
| 01 | Managed serviceMonthly fee, scoped to rollout | Scoped |
| 02 | GLM-5.3-Flash compute2,310 M tokens at published rates, no markup | ₹12,400 |
| 03 | Cap set by youUsage paused above this amount | ₹15,000 |
Different work consumes compute differently. Answer a few concrete questions and the planner returns a deliberately cautious monthly GLM budget, not a best-case teaser.
Default model · Managed GPU inference
Strong at coding, tool use and long documents, with a 1M-token context window. You pay for tokens processed, itemised at the published rates on the right.
Prompt
$0.45 / 1M tokens
Cached prompt
$0.09 / 1M tokens
Completion
$1.50 / 1M tokens
Question 1 of 2
Choose everything that applies. Human conversations, coding loops and unattended agents are kept separate because one “heavy user” label hides too much.
Most companies begin with the pilot, keep the managed service and move to dedicated infrastructure only when the workload calls for it.
Option
Test one workflow with one team before committing to a rollout.
Core service
Keep the assistant running after the pilot and extend it to more departments.
Option
For workloads that need reserved capacity, a chosen region or isolated processing.
We run the GPUs and run the model on them, on cloud infrastructure Qila manages. Nothing goes to OpenAI or a consumer AI service. The assistant application can also be hosted inside your own infrastructure so documents and history stay with you. Reserved capacity or a specific region is scoped as a dedicated deployment.
No. It returns the answer and retains neither the prompt nor the response, and prompts are not used for training. Your workspace may keep chat history if your company chooses to. Both policies are documented in writing before the pilot.
Yes. Limits are set by company, department or employee. When a limit is reached the workspace pauses access until an administrator changes it.
A low estimate is useless if the company hits its cap mid-month. The planner assumes a busy month, conservative cache rates and a 35% reserve. After the pilot we replace those assumptions with your measured cost per task.
GLM-5.3-Flash prompt, cached-prompt and completion tokens at published rates. It excludes the pilot and management fees, storage, search tools, document parsing, taxes and reserved GPU capacity.
Those sell per-seat software on infrastructure you do not control, and the bill grows with headcount. Qila runs the GPUs and the model, connects approved documents, gives you access and budget controls, and supports employees. You control the spend; we optimise the cost, so a firm of 25 or more usually pays less than it would for individual ChatGPT subscriptions.
Two to four weeks, depending on the document source, the access requirements and the workflow you choose.
Next step