The scheme runs its own compute, and routes to every brand.
TokenOne Delivery is the smart-routing layer underneath your wallet · quality-first delivery across every supported model and provider. Compute on TokenOne AI Apex / Core / Edge / Lite, or bring Anthropic / OpenAI / Google / AWS / Mistral / Cohere / Fireworks / Together via BYOK. Same wallet, same governance, every call.
TokenOne AI alongside the brands you already trust.
TokenOne Delivery is a peer to the major AI brands · not a layer above them. Pick the brand and the tier; TokenOne handles the wallet, governance and audit underneath.
TokenOne AI
Apex · Core · Edge · Lite
Native
Anthropic
Opus · Sonnet · Haiku
OpenAI
GPT-5 · GPT-4o · GPT-4o-mini
Gemini Pro · Flash
AWS Bedrock
Nova · Titan · partner pool
Mistral
Large · Medium · Small
Cohere
Command R+ · Command R
OpenRouter
Aggregator pool
TokenOne AI tiers are powered by leading open-source compute, packaged into four named tiers so the customer-facing unit stays simple. The underlying compute can swap underneath without changing the tier customers buy.
Four tiers. Pick the one that fits the workload.
Same logic as Anthropic Opus / Sonnet / Haiku · the tier is the unit you buy. The underlying compute is open-source, governed, sovereign-deployable, and swappable.
Apex
Highest reasoning. Multi-step synthesis, agentic chains, complex tool use. Frontier open-source compute.
Core
Balanced workhorse · quality + cost in equilibrium. The default for most production workloads.
Edge
Speed-first. Low-latency completions, classifications, summaries at scale.
Lite
Cheapest tier · sandbox, drafts, exploratory traffic. Free-tier eligible.
Three decisions on every call.
Every prompt is matched to a workload pattern, the right tier is selected, and the call is dispatched to the chosen brand. The decisions are deterministic, traceable and replayable.
Tier matching
Pattern matcher + classifier picks the right tier · Apex / Core / Edge / Lite · for the workload. Optimised for fit, not just spend.
Brand + endpoint
Within the chosen tier, the right brand + endpoint is selected on health, cost, residency and BYOK availability.
Adaptive feedback
Quality + latency outcomes feed back into the next decision. Drift detection escalates patterns that regress.
What TokenOne Delivery gives you.
Peer-brand catalogue
TokenOne AI alongside Anthropic, OpenAI, Google, Bedrock, Mistral, Cohere, OpenRouter. One brand list · your choice.
Four named tiers
Apex / Core / Edge / Lite. The unit customers buy. Underlying compute swaps underneath without breaking the contract.
BYOK throughout
Your provider keys plug straight in. Same governance, no rewrap, no vendor middleman.
Per-pattern policy
Define quality thresholds, latency targets, residency rules and brand allow-lists per workload pattern.
Six protocol tiers
Proxy, MCP, ACP, browser extension, CLI installer, advisory mode. One core, six surfaces.
Replay and audit
Every compute decision is a row. Replay against a new policy version to test changes safely.
The before / after on every call.
- No more "which model do we use?" · TokenOne Delivery picks the tier per workload pattern.
- No more "did the cheap path regress?" · quality validation runs every call.
- No more "we paid for what?" · every decision logged with cost and outcome.
- No more "can we switch brands?" · BYOK throughout, no vendor lock-in.
- No more "the platform team owns this" · the unit you buy is a tier, not a vendor SKU.