AI compute economics · measured, not quoted

The price of intelligence, measured daily.

Mintok prices AI compute the way markets price everything else: dollars per million tokens, computed from live model, chip, and rack data.

Self-Service

Buy the answer.

Benchmarks on your prompts, sizing reports, rack teardowns, agent designs — produced by agents, delivered in minutes.

From $99 · no sales call · sample reports below

Browse products →

Platform

Run the practice.

The workspace forward-deployed engineers run with their customers: size the fleet, plan orders, deliver against supply, monitor drift.

Invite-first · Free, Individual & Enterprise seats

See the platform →
Models · live
Kimi K2.6Moonshot · T6Kimi K2.5Moonshot · T5Kimi K3Moonshot · T6GPT-4o miniOpenAI · T4Llama 3.2 3BMeta · T1Llama 3.2 1BMeta · T1Gemma 2 2BGoogle · T1Qwen2.5 3BAlibaba · T1Claude Opus 4.7Anthropic · T7Grok 4.3xAI · T5Claude Sonnet 4.6Anthropic · T6Claude Opus 4.8Anthropic · T7Qwen 3 30B (MoE)Alibaba · T2Mixtral 8x22B (MoE)Mistral · T4
Silicon · live
Rubin Ultra VR300NVIDIARubin VR200NVIDIAMI500AMDMI455XAMDAscend 950HuaweiAscend 920HuaweiMTIA v3MetaMTIA v2MetaMaia 300MicrosoftMaia 200MicrosoftTrainium3AWSTPU v6pGoogle

Buy the answer

Deliverables, not demos.

Benchmark

$99

Your prompts across the leading models — measured cost, speed, judged quality. The cheapest model that clears your bar, highlighted.

View a sample report →

Instant Sizing Report

$699

Describe your workload in English; get sized hardware options with TCO, $/M-token, and the build-vs-buy break-even.

View a sample report →

Rack Teardown

$499

Every BOM line of an AI rack priced — cost pools, margin vs list, and who supplies each dollar.

View a sample report →

Eleven products, three subscriptions, two bundles — see the full catalog.

Run the practice

One workspace from first sizing to delivered racks.

01

Size

Every silicon vs every workload, in $/M-token.

02

Plan

Sized plans become orders with ATP.

03

Deliver

Supply ledger, POs, allocation lifecycle.

04

Monitor

Agent watchers flag drift from plan.

How the platform works →

The AI Compute Cost Index

The monthly index of what intelligence costs — rack build-costs, component sub-indices, hosted model price curves. $149/mo · Analyst Annual $1,490/yr.

Read Issue #0 →
◆ One deterministic kernel behind every number◆ Tenant isolation enforced in Postgres (RLS)Methodology & security published

Two ways in. Both start free.