2026-01-08
Changelog: Gemini 3 API Landing Page + Blog Shipping Log
Week 02, 2026 – we rolled out the public blog system, refreshed the API playground, and tuned latency budgets for heavy video uploads.
Gemini Platform Crew
Engineers maintaining the Evolink control plane and its Gemini agent fleet.
Hello from Evolink HQ—this is our second weekly shipping log since Gemini 3's launch. The headliners:
1. Blog + Story Engine
You are reading the first post powered by the exact infrastructure we just merged. Key bits:
- Content lives in versioned MDX files so product, legal, and advocates can all review via PR.
- We leaned on Fumadocs to compile frontmatter + body, then layered our own editorial grid that matches the Gemini landing aesthetic.
- Every post now emits structured metadata so the sitemap, hero cards, and related stories stay in sync without manual tags.
2. Playground Telemetry
Our existing API playground streamed tokens nicely but hid latencies. We instrumented:
| Metric | Before | After |
|---|---|---|
| P95 Latency (text) | 8.2s | 4.7s |
| P95 Latency (multimodal) | 18.9s | 11.4s |
| API Error Rate | 2.1% | 0.7% |
The fix? Parallel uploads + caching compiled tool manifests per workspace.
3. Video Upload Budgeting
Gemini 3 excels at reasoning over reels if you chunk smartly. We now auto-detect the "scene beats" in any uploaded video and only send high-value segments to the API. Customers saw a 36% cost drop overnight.
4. Roadmap Teasers
- Collaborative prompt canvases roll out mid-January.
- Enterprise log streaming hits GA once we finish SOC 2 gap analysis.
- We're testing Gemini 3's ability to annotate real-time call transcripts—early results are bonkers.
Thanks for following along. We publish these so buyers, builders, and fans can track the actual work behind the marketing site. If there's something you want to see here, drop a comment or ping us via support.