The headline is not who won - it is how the leaderboard has flattened. A year ago a single model typically held 40-50% of Vercel traffic. The April 24 snapshot shows no model above 24%, with open-weight Kimi K2.6 sitting second and Claude Sonnet 4.6 still earning a top-three slot despite being the older Sonnet generation.
For founders building agents and AI features, this confirms that model routing has become the default architecture. Pinning to one provider is now a cost and latency choice, not a quality one. Gemini 3 Flash dominates on throughput-heavy, low-complexity work; Kimi K2.6 pulls share on reasoning; and Claude still wins on code generation and long-context agents.
If you are still hard-coded to a single model, move to a router (Vercel AI Gateway, OpenRouter, or your own) this quarter. The fragmentation is now structural, not noise.
