Model LLM consumption at organizational scale — with heavy-tail user distribution, workload-profile decomposition, agentic amplification, and governance levers. Built for CIOs, CFOs, and infrastructure operators who need to estimate order-of-magnitude token spend before the invoice arrives.
On the defaults. Every number in this tool is a starting point. The tool's value is in letting you replace defaults with your own telemetry. Compute runs server-side; methodology is proprietary to SVAG Labs.
Organization Preset ready
Three reference configurations. Select one as a starting point, then edit any value.
Organization Shape
Workload Profiles
Seven canonical work patterns. Defaults are computed server-side; edit any field to override for your environment.
Tier Composition
What fraction of each tier's time goes to each workload profile. Rows should sum to 100%.
Model Pricing
Per-million-token rates across three model tiers. Defaults are Anthropic published list prices as of September 2026. Pick a model to load its rates, or type your own negotiated or other-vendor rates — editing any rate switches that tier to Custom.
Frontier Opus / Fable
Mid Sonnet
Cheap Haiku
Source: Anthropic API pricing documentation, checked 21 September 2026. Cache = cache-read rate (10% of input; 2.5% on Fable 5.1). Rates exclude the Batch API discount (−50%), US-only inference (+10%), and cache-write charges. Anthropic notes that Claude 4.7 and later models use a newer tokenizer that produces roughly 30% more tokens for the same text — if your token counts were measured on an older model, raise the per-call token inputs accordingly.
Governance Levers
Per-user monthly token cap0 = no cap.
Frontier → Mid routing (%)Share of frontier traffic routed to mid-tier (for example Opus → Sonnet).
Agent session timeout (hours)Caps amplification for agentic profiles. 0 = no cap.