Beta
Tokonomix is in open beta
We build in the open. This page lists — honestly — what works today, what is partial, and what does not exist yet. Use Tokonomix at your own risk during this phase.
What beta means here
- Use at your own risk — we do not offer a service-level agreement (SLA, a guaranteed uptime/response commitment) during beta.
- Your balance is safe — we never take or reset customer credits.
- Prices shown on the pricing page may change once beta ends — nothing is locked in yet.
- During beta we charge no per-call council fee. The token usage of every model in a council is billed per token, just like a direct call.
Works today
- OpenAI-compatible and Anthropic-native gateway endpoints — point your existing client at Tokonomix without rewriting it.
- A model catalogue of 100+ models across providers, kept up to date.
- Council modes — consensus, diff, best-of and full — with an independent judge model reconciling the answers.
- EU-only routing per API key, for requests that must stay inside the EU.
- Image generation and editing, plus text embeddings, through the same gateway.
- Per-key budget caps, so a single key cannot overspend.
- A live billing dashboard that shows exactly what each call cost.
- Speech-to-text transcription through the same gateway.
- Large-context staged upload is now consumable by the council — stage it once and every proposer and judge reads the same in-region context pack. And if you send grounding the council can't use, you get a clear signal instead of a silent, ungrounded answer. Inline context in the request body works too.
Beta / partial
- Tool calls inside streaming responses are buffered: they arrive all at once instead of token-by-token.
- A grounding "ask-back" feature exists — the model can ask a clarifying question before answering — but it is switched off pending calibration.
- Treating "the model can't answer" as a valid outcome (abstention) is still in development.
- Our own benchmarks show the council ties the best single model — it does not (yet) beat it on accuracy. See the benchmark numbers →
Not yet available
- Webhooks for asynchronous notifications.
- A formal service-level agreement (SLA).
- A marketplace for outsourcing jobs to other agents — in development.
Beta billing, in short
No per-call council fee during beta. The token usage of every council member (and the judge) is billed per token, exactly like direct calls. Single-model (raw) calls are billed as before. Your balance is never taken or reset.