What we've shipped lately
Model launches, pricing tweaks, dashboard upgrades and platform improvements — all in one timeline.
- Pricing
Cache prices now follow each vendor's own ratios
Cache prices now follow each vendor's own ratios, at the model's discount. Cache reads bill at 10% of input on every vendor (OpenAI and Gemini were 20%; Claude Fable 5.1 reads at 2.5%). Cache writes bill at 1.25× input on Claude and GPT-5.6+ (they were the plain input rate). OpenAI's long-context tier past 272K input tokens and Gemini 3.1 Pro's past 200K now apply: the whole request bills at 2× input and 1.5× output.
- Auth
New in-house auth (auth-v2)
Replaced Auth.js with an in-house JWT + magic-link + Google OAuth flow. Dashboard, login, and middleware all migrated.
- Platform
UserSay feedback widget installed
First-use trigger, identify hook and reward webhook all wired up across the dashboard.
- Pricing
Cache pricing improvements
Dropped the cache-write surcharge entirely; cache reads on OpenAI/Gemini dropped from 0.25× → 0.20× of input. Repeat prompts cost less.
- Dashboard
AI Agent tab on the homepage hero
Drop one paste-ready instruction snippet straight into your agent to get it calling Kunavo without writing SDK boilerplate.
- Platform
Prompt caching, end-to-end
Cache savings surfaced across the dashboard, marketing and docs. Affinity routing keeps repeated prompts warm on the same upstream node, up to 90% off input.
- Platform
SLO dashboard with TTFB and disconnect monitoring
Per-dispatcher TTFB and mid-stream disconnect rates now visible in the admin console. Easier to spot upstream regressions before users do.
- Platform
Public uploads on files.kunavo.com
User-uploaded images for image-edit and image-to-video are now served from files.kunavo.com — permanent URLs, neutral bucket, dedicated credentials.
- Platform
Balance reservation prevents concurrent overdraw
Estimated call cost is now reserved up front, so multiple concurrent calls can't drain an account past zero. Covered by a new concurrency test.