Model releases, price changes and platform updates.
All pricing is now in USD and matches each provider's public rate card exactly, with zero model markup. Revenue comes solely from a 10% top-up fee. Bill lines can be expanded to show the input / cache-read / output cost breakdown, the peak-window flag and the request_id, and export to CSV.
Input tokens that hit an upstream prompt cache are billed at the cache rate (as low as 1/50 of standard) and the saving is passed to you in full. DeepSeek peak windows are billed by the actual request time.
Added deepseek-flash (V4.1 Flash, 1M context), qwen-flash, qwen-plus and qwen-max; retired deepseek-coder, which the provider no longer offers. Auto routing tiers remapped in ascending cost order.
New product site and API documentation published; the merchant console entered development. Auto routing (complexity tiers) is generally available.
New "auto" model: routes to SIMPLE / MEDIUM / COMPLEX / REASONING tiers by request complexity, cutting cost on simple traffic by up to 95%.
Browser-based chat with streaming output, Markdown rendering, per-model pricing display and a live balance panel.
Completed the LiteLLM → AIGateway rebrand; ICP filing approved and service migrated to the new domain.