gawk.devgawk.dev AI Genesis Report · April 2026Published 2026-05-05
Who's actually winning in AI tooling — April 2026
30 days of verified data across models, SDKs, and tools. Every number links to its source.
Tencent Hy3 Preview climbed 25 OpenRouter ranks to #1 in April
Three Chinese open-weight models entered the top-25 while two OpenAI models fell out of the top-40.
Verify: openrouter.ai/rankings ↗
Climbing the OpenRouter ranks
Rank changes between the start and end of the window. Climbers only — see the next section for fallers.
OpenRouter request volume reflects developer API spending, not end-user adoption. Biased toward API-first workflows that route through OpenRouter — direct OpenAI / Anthropic / Google customers who never use OpenRouter are invisible.
- meta/muse-spark-1.3-contributor
- openrouter/auto
- google/gemini-3.8-flash
Falling on OpenRouter
Same window, opposite direction. Movement here is request-volume movement, not capability change.
OpenRouter request volume reflects developer API spending, not end-user adoption. Biased toward API-first workflows that route through OpenRouter — direct OpenAI / Anthropic / Google customers who never use OpenRouter are invisible.
- xiaomi/mimo-v2.5
- openai/gpt-5.6-luna
- deepseek/deepseek-v4-pro
SDK download gainers
Where developers are placing real install bets, ranked by % growth across PyPI / npm / crates / Docker / brew / VS Code.
- burn
- tch
- candle-core
SDK download losers
Same registries, packages losing developer base. Steepest decline first.
- Codeium.codeiumDaily values are derived from day-over-day diffs of the cumulative install counter exposed by Microsoft's marketplace catalogue. install ≠ active use — auto-installed bundle extensions, CI runners, and codespace pre-warms inflate the absolute number, but the WoW Δ signal is consistent.
- TabNine.tabnine-vscodeDaily values are derived from day-over-day diffs of the cumulative install counter exposed by Microsoft's marketplace catalogue. install ≠ active use — auto-installed bundle extensions, CI runners, and codespace pre-warms inflate the absolute number, but the WoW Δ signal is consistent.
- llamaindex
AI labs by GitHub activity
Ranked by total public-event volume across each lab's flagship repos. Geographic spread visible per row.
- Anthropic · San Francisco, US200 eventsanthropic.com ↗
- LangChain · San Francisco, US196 eventslangchain.com ↗
- OpenAI · San Francisco, US188 eventsopenai.com ↗
- Microsoft Research · Redmond, US183 eventsmicrosoft.com ↗
- Hugging Face · New York, US177 eventshuggingface.co ↗
Provider incident-days
Active incidents on the public status pages, summed per UTC day captured. More incident-days = more days a provider was flapping.
- openai-api9 incident-daysgawk.dev sources ↗
- claude-code7 incident-daysgawk.dev sources ↗
- copilot1 incident-daygithubstatus.com ↗
Subscribe to the daily digest
The same lens — every number cites its source, no scoring, no editorialised takes — applied every UTC day. We don’t promise more reports like this one; the daily is the cadence.
Subscribe →