Changelog

Platform changes, new model launches, API version records. Newest first.

Updated 2026-08-03

2026-08-03

🔧 Service disruption notice (06:09–06:37 UTC, resolved) A TLS certificate auto-renewal failure interrupted API calls and console access for 28 minutes. The certificate has been reissued (valid until 2026-11-01) and affected customers have received account credit. Certificate expiry alerting and automatic reloading on certificate change are now in place

⚠️ [Breaking · effective 2026-07-24] SLA timeout threshold changes from 60s to 90s The “upstream did not return the first byte” threshold moves from 60 to 90 seconds. The original 60s was written without checking it against the implementation, and the gateway never actually enforced it. This change also makes the threshold enforceable for the first time (a first-byte bound now covers both streaming and non-streaming requests). Measured over the last 30 days: 0.0293% of requests exceeded 60s to first byte, 0.0052% exceeded 90s

Image Playground launched The console adds an image generation / editing Playground that calls /v1/images/* with your real API Key, billed as usual
Top-up receipt export The billing page can export payment receipts (PDF); billing-profile management added, with a guided fill-in before first export

Per-provider call examples on the pricing page Claude models show a Claude Code setup guide; OpenAI models get curl examples

Promo codes for top-ups Top-up checkout now accepts promo codes, applied by type / amount / threshold / usage limits

Onboarding upgrade New-user guide widget + setup tutorials (beginner 0→1 and expert fast paths); registration no longer force-redirects to the top-up page
cc-switch one-click import Claude Code / Codex / Gemini CLI configs can be imported via cc-switch deeplink — no manual env-var editing

Top-up bonus campaigns First-charge bonus and a general top-up bonus engine are live; the top-up dialog shows bonus and credited amounts directly
Affiliate program launched Distribution center is live: personal referral links, tiered commission, commission wallet (convert to balance or withdraw via USDT)
Pricing page upgrade Explicit USD billing with CNY reference prices

Transparency Center launched /status is open: service availability monitoring (real Uptime Kuma data) plus account-pool notes

Site available in 5 languages Chinese / English / Russian / Japanese / Vietnamese, with IP-based language routing

Blog launched /blog is now open, with a first batch of 3 posts: dual-channel isolation reliability, a Prompt Cache cost benchmark, and refund rules as product
Docs filled out Four formal docs are live: Quickstart / API Reference / FAQ / Changelog
Nav redesign The public site nav adds a Blog entry, reordered to Models / Pricing / Docs / Status / Blog

Epic 4 done All 9 stories of the AI API proxy engine shipped: OpenAI / Anthropic dual-format passthrough, dual-channel isolation, automatic upstream failover
Epic 6 done 13 console stories shipped: real-time usage over WebSocket, a three-state home, API key management, usage history, first-call onboarding, and more

Low-balance email alerts An email is sent automatically when the balance drops below the threshold (default $5); can be turned off under Notification preferences in the console
WebSocket connection limit Up to 5 concurrent WS connections per account; the client is clearly notified on overage (no silent drops)

Dedicated SMTP sender Platform notification emails are sent from noreply@zetaapi.ai, with a customizable SMTP_FROM
Switched to Resend Email moved from a self-hosted Postfix to Resend, raising delivery from 91% to 99.2%

Dual-channel isolation launched Each model has a primary + fallback as two independent channels; when the primary stays below 95% availability for 5 minutes it auto-switches to the fallback
Compensation rules launched Published the payout algorithm and tiers; incident-window compensation is settled to your balance after a reviewed claim (correction: this entry previously described compensation as fully automatic, which was ahead of the implementation)

Anthropic native format support The /v1/messages endpoint is live, the anthropic-version header is passed through, and prompt cache cache_control is fully compatible
Claude Sonnet 4.5 launched at a 0.6x multiplier

One-line install CLI npx zetaapi setup --api-url=https://api.staging.cosmos-router.com claude-code writes ~/.claude/settings.json automatically; setup cursor prints the values for manual entry (correction: this entry previously implied cursor was automatic too)

Public beta launched The api.zetaapi.ai domain opened for registration; the first batch supports five model families: Claude / GPT / Gemini / DeepSeek / Qwen

Breaking changes are marked with ⚠️ and announced in-console at least 14 days ahead
New model launches don’t affect existing APIs; retired models keep at least a 90-day transition window
Subscribe via RSS: /blog/rss.xml (coming soon)