Unlock agentic.
Fix the bill
Agent Libre replaces the major AI providers' usage-based billing with a fixed plan based on throughput. Your teams and your AI agents work without counting tokens; your finance team knows the amount in advance.
An honest and transparent monthly plan
Free 24-hour to one-week audit. No commitment.
cat > ~/.claude/settings.json << 'EOF' { "env": { "ANTHROPIC_BASE_URL": "https://api.agentlibre.ca", "ANTHROPIC_AUTH_TOKEN": "al_live_••••••••", "ANTHROPIC_MODEL": "glm5.2-agentlibre" } } EOF
Usage-based pricing punishes ambition
You've reached your usage limit. Try again after your limit resets.
A serious agentic workflow consumes relentlessly. Billed by the token, it becomes impossible to budget for — and the organization's first instinct is to limit it.
Our mission: Unlock Agentic.
A fleet of AI agents running continuously burns through a quarter's budget in three weeks. No one dares run it in production.
Per-seat quotas, downgraded models, truncated context: you end up optimizing the bill instead of the product.
Spend swings by 300% from one month to the next. You can't commit to an annual budget on that basis.
High-volume use cases — bulk analysis, autonomous AI agents — stay stuck at the slide-deck stage.
You stop buying tokens.
You hand over your inference.
A managed service provider (MSP) runs your mailbox, your website or your security systems. Agent Libre does the same for your AI inference: capacity sized, operations watched, security, billing. Neither a model provider nor a reseller — we run your inference layer, in Canada.
- You buy tokens by the unit
- You absorb every swing in load
- You watch quotas and rate limits
- Your provider decides the model and the context
- You are exposed to price hikes and the dollar
- You reserve throughput, not tokens
- Variation is measured; the rate holds
- We run capacity and availability
- We syndicate load across members
- You cut out every middleman
- You are assured the best cost
What you get.
- Guaranteed throughput, reserved up front rather than metered per use
- Continuous model optimization on the GPU server
- Upgrades to newer models (GLM 5.2 to 5.3, Kimi)
- On the same hardware, available throughput rises over time
Your infrastructure, no middleman.
- Initial matchmaking of syndicate members
- Departures, arrivals and throughput reallocation
- Optimization frees capacity: more members on the same GPUs
- Data-centre partners, lease negotiation and renewal
The syndicate's volume works for you.
- Buying power that grows with the members' combined volume
- Continuous tracking of the GPU market
- Renegotiation of GPU agreements every 6 to 12 months
- Sale or donation of unused throughput
That is what the flat rate covers: the operations, the administration and the negotiation.
What your workload costs with us
Enter your workload, then compare it with the pay-per-use cost of proprietary models. The exact plan is confirmed after 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.
Compare in detail
Not all tokens are equal. Input (your context) is read in parallel — the hardware processes hundreds of thousands of tokens per second. Output is generated one token at a time, and each token requires a full pass through the model. Producing a token genuinely costs several times more than reading one — which is why usage-based providers price output 3–5× higher than input.
In agentic workloads, input dominates: every call re-reads the full history — several thousand tokens read for every thousand generated. The prefix cache makes that re-read much lighter: one PTU processes 3,205 cached tokens per second, versus 595 uncached. Anthropic bills input at $3 US/M (~$4.11 CAD) and Z.ai at $1.40 US/M (~$1.92 CAD). Within your PTUs, at full utilization, it comes to about $0.92 CAD/M uncached and $0.17 CAD/M from the cache.
| Anthropic (Sonnet)* | Z.ai (GLM 5.2)* | Alibaba Cloud (GLM 5.2)* | Agent Libre (GLM 5.2) | |
|---|---|---|---|---|
| Input | $3 US/M (~$4.11 CAD) | $1.40 US/M (~$1.92 CAD) | $0.97 US/M (~$1.32 CAD) | ~$0.92 CAD/M uncached; ~$0.17 cached† |
| Output | $15 US/M (~$20.55 CAD) | $4.40 US/M (~$6.03 CAD) | $3.04 US/M (~$4.16 CAD) | ~$2.89 CAD/M† |
| Long context | up to 1M tokens | up to 1M tokens | up to 1M tokens | up to 1M tokens per request, counted in PTUs‡ |
* September 2026 prices, at regular price with no promotion.
† Full-utilization equivalent, at $1,438 per PTU (12 PTUs and up). The plan stays fixed: you pay for your PTUs, not for the tokens consumed. One PTU processes about 189 tok/s of output, or 595 tok/s of input, or 3,205 tok/s of cached input.
‡ A longer request counts more input tokens, hence more PTUs. Beyond your PTUs, requests go through on a best-effort basis, with no throughput guarantee.
GLM 5.2 supports 1M tokens of context natively, and its sparse attention keeps processing cost from climbing steeply as context grows. A longer request simply counts more input tokens in your PTU sizing. What sets us apart from usage billing is what happens between turns: an AI agent re-reads its entire history on every call, and a usage-based provider re-bills that re-read in full, every time. Here, your PTUs are fixed: re-reading the context uses reserved capacity, not a bill that keeps climbing.
The plan is confirmed by your real consumption
The calculator above gives you a ballpark. Before committing to anything, the 24-hour to one-week audit measures your real consumption with your current provider and compares it to what a fixed-throughput plan would cost.
We plug a gateway in front of your current provider, no commitment. Your teams work as usual; we measure everything: throughput, concurrency, peaks.
The plan we propose reflects the load observed during the audit.
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.
“The Canadian alternative to OpenRouter”
Understand your AI bill before it decides for you
Cost structure, impact on your margins, a five-year projection and the risk of depending on a single provider.
- The real structure and dynamics of your AI bill
- Its impact on your margins and your competitiveness
- Five-year projection: costs and usage are rising — what about you?
- Betting everything on a single provider: the concrete risks
- AI is now part of your operating margin — and between geopolitics, the US dollar and tariffs, who controls it?
The model behind Agent Libre
Our gateway serves GLM 5.2, Z.ai's open-weight model. Here's how it stacks up against the leading proprietary models, based on public, independent evaluations.
| Benchmark | GLM 5.2 | Claude Sonnet 5 * | Claude Opus 4.8 | GPT-5.5 | What it measures |
|---|---|---|---|---|---|
| SWE-bench Pro | 62.1 | 63.2 | 69.2 | 58.6 | Real fixes across 41 professional repos |
| Terminal-Bench 2.1 | 81.0 | 80.4 | 85.0 | 84.0 | Agentic command-line tasks |
| FrontierSWE | 74.4 | — | 75.1 | 72.6 | Frontier software engineering |
| IDOR detection (Semgrep) | 39% F1 | — | 32% F1 | — | Application security, no scaffolding (vs Claude Code) |
More detail
Bottom line: Opus 4.8 keeps the edge on the longest repo-scale tasks, and GLM 5.2 comes within a point of it on FrontierSWE while beating it on scaffold-free vulnerability detection. For the bulk of day-to-day agentic work, the quality gap is marginal — the cost gap isn't. That's what makes the fixed plan possible.
The underlying trend works in our favor: open-weight models are closing in on the private labs by the month. Kimi K3, released open-weight on July 27, 2026 (2.8 trillion parameters, 1M-token context), ranks 3rd on Artificial Analysis's intelligence index — behind only Claude Fable 5 and GPT-5.6 Sol Max — and 1st on the Frontend Code Arena. The gap between the best model you can rent and the best you can run yourself has shrunk from six-to-nine months to three-to-five; one Kimi generation closed thirteen index points in three months.
No rewrite
Agent Libre speaks your current providers' protocol. Your integrations keep working exactly as they are.
What you build around the model stays yours.
Your gateway stays in charge ; we supply the throughput behind it.
Security and protected data
Infrastructure and operations in Canada.
TLS 1.3 in transit, AES-256 at rest, managed and rotated keys.
Prompt analysis and optimization, protection against sensitive-data leaks and token theft, no model training.
Learn moreOur infrastructure providers hold SOC 2 Type II attestation, with ISO 27001 / 27017 and GDPR or HIPAA compliance depending on the provider. Exportable audit logs.
A free
24-hour to one-week audit
We plug a gateway in front of your current provider for 24 hours to one week — nothing changes in your tooling — and hand you a consumption report compared to your current bill.
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.