Experiential Labs Pricing 2026 — Free Gateway, 0% Token Markup

What the free plan includes

The hosted gateway exposes an OpenAI-compatible API for hosted, customer-key, and local models. The official plan includes model routing, budgets, allowlists, attribution, and community support. Teams may mix their own keys with platform credits.

Read our Experiential launch and pricing analysis for the router evidence, fine-tuning boundary, and buyer test.

Costs the $0 price does not remove

BYOK requests remain payable to the connected provider. Credit-funded calls stop when the buyer's cap or balance is exhausted. Self-hosters pay for machines, storage, monitoring, upgrades, and the models they serve. Router optimization can also use embedding, simulation, judge, and candidate-model calls; supervised fine-tuning adds training and serving cost.

Compare current upstream rates in the AI API pricing table or estimate a fixed workload with the token cost calculator.

Why there is no Experiential model rate

Experiential is a gateway, not a single inference SKU. One alias can resolve to a first-party API, OpenRouter, another OpenAI-compatible provider, a local deployment, or a fitted project router. The payable unit rate therefore follows the exact upstream route and customer configuration.

The canonical pricing dataset and public pricing API contain zero Experiential model rows. We will not represent free gateway software as free model inference.

Labs availability and reproducibility blocker

Experiential cannot receive one model score in AI Pricing Guru Labs. A useful test must pin the gateway and catalog, split real traffic into fitting and held-out sets, preserve route decisions, and measure accepted-task quality, token usage, latency, retries, failures, and billed cost. A trained-model test also needs the exact base model, training schedule, artifact, serving price, and paired held-out outputs.

The current release evidence uses deterministic simulation, did not run credential-gated live provider calls, and did not produce a paid trained artifact. Labs therefore publishes a blocker instead of converting project evidence into a live leaderboard result.

What to verify before adoption

• Which requests use BYOK, platform credits, or local compute.
• Whether every alias exposes a verified model identity and price.
• How routing changes are approved, versioned, and rolled back.
• Whether quality labels come from an independent acceptance test.
• The full optimizer and fine-tuning budget, not only inference tokens.