The Saanora Platform is live, with Clara
clara-1 is the first model on the platform: a frontier decision model that answers typed questions (yes/no, choice and score) about any record, with a calibrated probability for every option. It scores 93.55% on AgentDecisionBench tool routing, level with Jev 1.13, and 83.6 on JevBench-mini coherence, level with the leaderboard's #1. Five decisions about one record take about 0.1 s.
Requests take an optional mode. fast, the default, reads the record once for all questions. consistent reads it again with each question, which keeps answers steadier when options change or a question is reworded. In both modes each question is answered on its own. POST /v1/systemone accepts the open System One format, so existing System One clients work by changing the base URL.
Pricing is per input token: $0.50 per million, with $0 output and no per-decision fee. Credit is prepaid in top-ups from $5 and never expires.
Usage tiers run from Tier 0, which is sign-up only with no API access, to Tier 5. Your first top-up unlocks the API at Tier 1, and requests and tokens per minute then rise automatically with lifetime purchases and the days since your first payment, up to 5,000 requests and 5M tokens a minute. The largest single top-up grows with your tier, from $25 to $1,000.
The console ships with organizations and projects, API keys, usage by model, project and key with CSV export, prepaid billing with receipts and invoices for every payment, members with roles, per-project monthly spend limits, and a history of every tier change.
Mark, the language model behind Saanora Chat, is listed as mark-1 and is coming to the platform on the same keys, credit and limits.