Your monthly plan,honest and transparent
The plan holds two amounts: the server cost, billed with no markup, and the Agent Libre fee. Both are calculated on the number of PTUs you reserve, sized on your peak workload. The amount does not vary with the volume your AI agents actually consume.
Join a growing community in sovereign AI.
Amounts in Canadian dollars, taxes extra.
PTUs, sized on your workload
You reserve a whole number of PTUs, provisioned throughput units. Each PTU is a fixed share of the server that processes a given volume of input, cached input or output tokens per second. A request consumes a mix of the three. The number of PTUs is calculated from your peak requests, your average request size and your cache hit rate. That is what the agreement reserves for you, for its full term. Its price, on the other hand, tends to go down over time.
There is no separate context to reserve: a longer request counts more input tokens, hence more PTUs. Requests are accepted up to 1M tokens of context. Beyond your PTUs, they go through on a best-effort basis, with no throughput guarantee.
| One PTU processes, either | Tokens per second |
|---|---|
| Uncached input | 595 |
| Input served from cache | 3,205 |
| Output | 189 |
PTU: provisioned throughput unit. PTUs required = requests per second × (uncached input ÷ 595 + cached input ÷ 3,205 + output ÷ 189), rounded up. The server holds about 41 in total, 37 of them sellable, to be shared among the members; the rest is held in reserve for peaks.
Your plan, by workload
Enter your workload, then compare it with the pay-per-use cost of proprietary models. Amounts follow the September 2026 grid; the number of PTUs you really need is established by the free 24-hour to one-week audit.
PTUs required = requests per second × (uncached input ÷ 595 + cached input ÷ 3,205 + output ÷ 189), rounded up. The minimum plan is 1 PTU. Each PTU carries the server cost with no markup, $1,126 ($41,650 ÷ 37 provisioned PTUs), plus the Agent Libre fee at the tier reached: $483 from 1 to 5 PTUs, $383 from 6 to 11, $312 from 12. * Pay-per-use cost, September 2026 prices: the same tokens at each provider’s published rates, over 43,800 minutes a month, converted at 1.37. The compared models are not equal in quality on every task: the comparison is the cost of the same volume. The exact plan is confirmed by the free audit.
Operations, billed at cost
The server carries real expenses every month. They are pooled and split among the members in proportion to their PTUs, with no markup. Here is what they cover, at the September 2026 reference values. These amounts are estimates: the real expenses are established when the server goes live and written into the agreement before it takes effect.
| Expense | What it covers | Per month | Share |
|---|---|---|---|
| GPU lease* | 8 NVIDIA B200 GPUs, Canadian colocation included | $36,000 | 86% |
| System administration** | Deployment, model updates, throughput tuning, on-call | $3,750 | 9% |
| Network and bandwidth | Outbound traffic, IP addresses, links to the data centre | $500 | 1% |
| Observability | Gateway, per-member measurements, exportable audit logs | $400 | 1% |
| Security and governance | TLS 1.3 and AES-256, key management and rotation, prompt protection | $300 | 1% |
| Outage fallback | Capacity bought elsewhere during an incident, about 10 GPU-days a year | $700 | 2% |
| Total | $41,650 | 100% |
* GPU lease: US$4.50 per GPU-hour for 8 B200s in Canadian colocation, on a 6 to 12 month commitment. The median on-demand market price in September 2026 sat around US$6.13, and the observed floor at US$3.75. Converted at 1.37 CAD/USD. If you buy your GPUs instead of leasing them, this line disappears: see buying GPUs.
** System administration: loaded annual cost of about $180,000 CAD, a quarter of which — $3,750 a month — is allocated to the server. Each member pays their share of it: $1,216 in the example below, which is 32.4%.
Those $41,650 spread over the 37 PTUs provisioned on the server come to $1,126 per PTU, per month. The 10% of capacity held back for peaks is included in that cost. Your share follows the number of PTUs you reserve.
The server cost is divided by the number of PTUs Agent Libre provisions on it. Every optimization that fits more PTUs on the same hardware — better throughput tuning, a more efficient model version — lowers the cost per PTU for every member. The Agent Libre fee stays set per PTU: the more PTUs Agent Libre provisions on the same server, the more it earns, and the lower your cost.
| PTUs provisioned on the server | Cost per PTU, no markup | 12 PTUs: server cost |
|---|---|---|
| 37 (today) | $1,126 | $13,512 |
| 40 | $1,041 | $12,495 |
| 45 | $926 | $11,107 |
The operations rate is frozen when the contract is signed, for your whole commitment.
A renegotiated GPU lease, better throughput tuning, a new member joining the server: each decrease lowers the rate and is passed on to you at renewal.
The outage provision that went unused is credited to the group on each anniversary of the server, and reduces the following year's operations by the same amount.
The Agent Libre fee
This is what Agent Libre earns, set per PTU. It comes down as the number of PTUs you reserve goes up, and does not depend on the server cost.
| PTUs reserved | Agent Libre fee | Total with server cost |
|---|---|---|
| 1 to 5 PTUs | $483 / PTU / month | $1,609 / PTU |
| 6 to 11 PTUs | $383 / PTU / month | $1,509 / PTU |
| 12 PTUs and up | $312 / PTU / month | $1,438 / PTU |
The total adds the server cost, $1,126 per PTU today. All your PTUs are billed at the tier reached. The minimum plan is 1 PTU, or $1,609 a month.
The Agent Libre fee covers assembling and managing the group that shares the server, negotiating and tracking the GPU leases, watching the market and migrating to new models.
This fee may be indexed by at most 5% on each anniversary of the server going live, with written notice 60 days ahead. Operations stay billed at cost, with no indexation.
12 PTUs for 1,100 tok/s of output
This member generates 1,100 output tokens per second at peak and re-reads 10 tokens for every token written, 85% of them from the cache. That takes 12 PTUs, which is 32.4% of the server's PTUs. That proportion applies to each of the group's expenses.
| Expense | Full server | Monthly amount |
|---|---|---|
| GPU lease | $36,000 | $11,676 |
| System administration | $3,750 | $1,216 |
| Network and bandwidth | $500 | $162 |
| Observability | $400 | $130 |
| Security and governance | $300 | $97 |
| Outage fallback | $700 | $227 |
| Operations | $41,650 | $13,512 |
| Agent Libre fee (12 × $312) | $3,744 | |
| Monthly plan | $17,256 |
At 30 tokens read for every token written, the same output throughput takes 23 PTUs, or $33,074 a month. Rounding can produce a difference of a few dollars between the sum of the expenses and the subtotal.
The same capacity, billed by usage
A throughput of 1,100 tok/s running without pause for a month generates 2.89 billion output tokens. Here is what those same tokens would cost elsewhere, at providers that serve GLM 5.2 in FP8 like our server and at Anthropic, at three utilization rates and two load profiles: 10 tokens read per token written, then 30 to 1. The prefix cache is assumed to hit 85% of the time, in line with what OpenRouter shows for the most used GLM 5.2 providers, 81% to 91%. At Agent Libre, these two profiles take 12 and 23 PTUs.
| Provider* | Read / written | 100% util. | 70% util. | 50% util. |
|---|---|---|---|---|
| Alibaba Cloud | 10 : 1 | $24,266 | $16,986 | $12,133 |
| Alibaba Cloud | 30 : 1 | $48,751 | $34,126 | $24,375 |
| Z.ai | 10 : 1 | $34,495 | $24,147 | $17,248 |
| Z.ai | 30 : 1 | $68,634 | $48,044 | $34,317 |
| Baseten Fast | 10 : 1 | $45,683 | $31,978 | $22,842 |
| Baseten Fast | 30 : 1 | $84,772 | $59,341 | $42,386 |
| Anthropic Sonnet 4.6 | 10 : 1 | $87,327 | $61,129 | $43,663 |
| Anthropic Sonnet 4.6 | 30 : 1 | $143,168 | $100,218 | $71,584 |
| Anthropic Opus 4.8 | 10 : 1 | $145,545 | $101,881 | $72,772 |
| Anthropic Opus 4.8 | 30 : 1 | $238,614 | $167,030 | $119,307 |
| Agent Libre | 10 : 1 · 12 PTUs | $17,256 | $17,256 resale: $14,918† donation: $16,328‡ | $17,256 resale: $13,360† donation: $15,709‡ |
| Agent Libre | 30 : 1 · 23 PTUs | $33,074 | $33,074 resale: $28,544† donation: $31,295‡ | $33,074 resale: $25,523† donation: $30,109‡ |
* September 2026 prices, at regular price with no promotion, as posted on OpenRouter.
† Resale, on our estimate of the September 2026 market: the unused share, 30% or 50% of your PTUs, is put up for sale on OpenRouter at US$0.70 per million uncached tokens read, $2.20 for output and $0.105 read from the cache, the low end of GLM 5.2 providers without a temporary discount. We assume half of that share finds a buyer and 5% in fees. The proceeds are credited on your plan; the actual credit depends on demand.
‡ Donation: the unused share goes to an eligible organization, valued at the cost of our GPUs, $973 per PTU per month. The tax saving uses the combined Québec and federal rate for a corporation at the general rate, 26.5%; an SME eligible for the small business deduction saves less, about 12.2%. This is not tax advice: see our philanthropy offer.
Every GLM 5.2 offer in the table runs in FP8, like our server: the comparison covers the same model, at the same precision, at regular price with no promotion. None guarantees throughput: they bill every token consumed. At Agent Libre, the amount is fixed: what is read, written or served from the cache is covered by your PTUs.
Against Alibaba Cloud, the lowest regular FP8 price, the plan comes out ahead above about 71% utilization at 10 tokens read per token written, or 68% at 30 to 1; 57% and 53% if the unused share is resold (†). Below that, pay-per-use costs less, with no guarantee. What reserved PTUs give you on top: throughput guaranteed at peak, governance over your AI traffic and a server in Canada.
See the published unit rates
| Provider (model)* | Read | Written | Read from cache | Guaranteed throughput |
|---|---|---|---|---|
| Alibaba Cloud (GLM 5.2, FP8) | $1.32 / M | $4.16 / M | $0.26 / M | No |
| Z.ai (GLM 5.2, FP8) | $1.92 / M | $6.03 / M | $0.36 / M | No |
| Baseten Fast (GLM 5.2, FP8) | $2.88 / M | $9.04 / M | $0.29 / M | No |
| Anthropic Sonnet 4.6 | $4.11 / M | $20.55 / M | $0.41 / M | No |
| Anthropic Opus 4.8 | $6.85 / M | $34.25 / M | $0.69 / M | No |
| Agent Libre (GLM 5.2, FP8) | $1,438 to $1,609 CAD per PTU, per month | Yes | ||
Amounts in US dollars are converted at 1.37. The comparison counts a month as 43,800 minutes, or 730 hours. At Anthropic, a cache read bills at a tenth of the input rate, and inference restricted to the United States adds 1.1 times. One Agent Libre PTU processes about 189 tok/s of output, or 595 tok/s of input, or 3,205 tok/s of cached input, and a request consumes a mix of the three. At $1,438 per PTU and full utilization, that comes to $2.89 per million tokens written, $0.92 per million uncached tokens read, or $0.17 per million read from the cache. Anthropic's models do not compare directly on quality across every task: the table shows the cost of usage billing for an equivalent volume. The tokenizer in Opus 4.7 and later produces up to 35% more tokens for the same text. * September 2026 prices: third-party rates are those posted on OpenRouter, at regular price, with no promotion. The table only keeps providers that serve GLM 5.2 in FP8, our server's precision: FP4 offers (DeepInfra, CoreWeave, Mistral, Parasail) process about twice as many tokens per GPU at reduced precision, and those whose precision isn't published (DigitalOcean, Together, Fireworks) are left out. These rates can change without notice.
From signature to go-live
The server goes live once its 37 PTUs are allocated under contract. Until then, you're not left waiting: your AI gateway is in place from signature, and your traffic is served through inference brokerage.
It centralizes your company's AI traffic, measures your throughput and gives you every benefit of enterprise governance.
Your requests go to third-party inference providers, the way OpenRouter does, in line with your AI policies, such as Canadian hosting.
Your traffic moves to the server you co-rent with your syndicate's members, managed by Agent Libre. Your 12- to 24-month commitment starts.
What you sign
12 to 24 months, starting when the server goes live. Written notice 90 days before the end date stops the renewal.
It is set with the other members and follows the GPU lease: the longer the commitment, the lower the GPU rate, and that gain comes back to you. The options are shown to you before the server goes live. We renegotiate the GPU price during the term: if it comes out ahead, penalty included, your server cost per PTU drops by the same amount.
Your commitment only starts running when the server goes live: see the waiting list steps. Withdrawing during that period costs one month of the plan. That amount keeps every member firmly committed, which is what allows the group to form.
Three months of the plan, on top of the amounts owed for services rendered until then. That amount covers the time it takes to reallocate your capacity to another member.
If Agent Libre fails to meet its service guarantees, you can end the agreement without penalty.
You can add PTUs at any time, subject to what is still available on the server; otherwise the additional PTUs go through inference brokerage until a new server goes live. A reduction partway through is treated as a partial ending, prorated to the PTUs released.
The plan is confirmed by your real consumption
This page gives an order of magnitude. Before any commitment, the free 24-hour to one-week audit puts a gateway in front of your current AI provider. Your teams work as usual during that time. We measure your peak requests, your request sizes and your cache hit rate, then translate them into PTUs and compare with what your current provider charges. The proposed plan reflects the load observed.
Join a growing community in sovereign AI.
The server fills up one member at a time. Book a 30-minute exploratory meeting: we work out how to calculate your capacity, offer you the matching plan, then install your AI gateway and route your traffic through inference brokerage until the server goes live.
Your company has two supply chains
The first one you already know. Materials come in, they are transported, stored, transformed, delivered. Every step has a supplier, a lead time, a unit cost. When the margin tightens, you walk down that chain step by step, and you generally know where to look.
The second one moves information instead of materials. A price request comes in and becomes a quote. An order comes in and becomes a production order, an invoice, a confirmed delivery. A site report becomes a change order, or it does not. This chain runs through every one of your departments and it employs a large share of your people.
Let’s talk about your AI bill.
Describe your AI agents, your orchestrator and your current bill.