[ INFERENCE, ON YOUR TERMS ]
Buy credits, use the models you need, and see where every dollar goes. One key. No mandatory subscription.
Text & code · Usage-based billing · Solana USDC or card
[ TRY A MODEL ]
Compare the same task, with the same model and token usage.
DeepSeek V4.1 Flash
Example walkthrough · no account charge
Illustrative estimates. Final cost depends on verified usage and the rate accepted for your request.
Usage breakdown below uses provider net rates.
Buyer total = $1.86 ÷ 0.90.
Platform fee = buyer total × 10%.
The platform fee is included in the buyer price. Values shown to six decimal places.
8 models in the catalogue · Output price / 1M tokens
Illustrative prices include the platform fee. Reference prices are public retail snapshots for the same route, dated 16 Sep 2026. Savings vary by model; input and cache are billed separately.
Compare models, capabilities and pricing in one place.
[ USAGE PATTERNS ]
Requests, tokens and model activity from a fixed public reference dataset.
Source: Engy public activity ↗. Captured 18 Sep 2026. The chart covers 18 Aug–16 Sep 2026, UTC. The current partial day and projected values are excluded.
Complete-day totals: 23,550,647 requests and 328,943,005,132 tokens. The source displayed 24,292,807 requests and 343,383,362,119 tokens under its own 30d window. This page consistently sums complete daily buckets instead.
Success rate and average total duration use the source snapshot’s separate 24h window. Duration is not time to first token. Reference models do not imply availability in the InferLi catalogue. This dataset never changes your account balance.
Your own history starts with your first call.
View my usage ↗02 / From choice to your first call
Add credits, create your personal key and start using your chosen model. Advanced controls stay available when you need them.
Review your amount and any payment fees before checkout.
Copy your key into an OpenAI-compatible tool or app.
Find every call and its cost in your account.
THE $AI ECOSYSTEM
$AI coordinates the ecosystem. Staking rewards are delivered as $CREDIT inference credits, which can be activated into AI usage balance.
No. You can add AI usage balance directly. Holding or staking $AI is a separate way to participate in the ecosystem.
Start with the exact model and required capabilities, then check provider terms and capacity. Lowest-cost routing compares the estimated whole request. Performance priority needs provider measurements. You can also pin a named provider; switching to a different model is not the default.
A rate limit or connection failure can try another eligible route within your price and data constraints. Once streaming has started, output from another provider is not spliced in. If the upstream may have executed but the response is lost, the request is reconciled instead of blindly retried. The final buyer bill is deduplicated by request ID.
The gateway processes request content and the selected provider receives it. Review each offer’s retention, training-use and region terms before sending sensitive work. Routine records are designed around model, usage, cost, latency and error metadata; troubleshooting content capture requires a separate opt-in.
