Kimi K2.6 Joins The Frontier Lineup, Auto-Router Opens Up

April 21, 2026

Moonshot's new Kimi K2.6 is live in the model picker as a frontier option. And Auto routing now picks from every model your plan can run, not just one default — so the match between your prompt and the model handling it is much tighter. Community goes from 1 model on auto to 11. Pro plans go from 6 to 14.

What's new

• Kimi K2.6 is selectable as a frontier-tier pick alongside Opus 4.7, GPT-5.4, and Sonnet 4.6. It bills at the workhorse rate, so it's accessible on Community and Pro alike.
• Auto routing actually uses your full pool now. Before, Community always landed on Gemini 3 Flash and Pro plans bounced between six options. Now Auto reads your prompt — its length, its tool-call shape, how deep it needs to go — and picks the best fit from every model your tier is allowed.

When you'd notice it

• You're on Community and ask for a long refactor plan. You used to land on Gemini 3 Flash regardless of the task. Now Auto sees the depth and reaches for Kimi K2.6 or Qwen 3.5 397B — still inside your budget.
• You're on Pro Max and paste a 200K-token research corpus. Auto now has the full premium shelf — Opus 4.7 for frontier reasoning, Kimi K2.6 for long code analysis, GLM 5.1 when reasoning depth matters more than polish.
• You're running an agentic workflow with 20+ tool calls. Auto sees the tool count and promotes to a frontier pick instead of grinding through the plan on a budget model.

Try it

• "Plan and run a deep refactor of this module — pick whichever model handles the task best."
• "Compare two reasoning paths through this brief in swarm mode."
• "Auto-route this and just give me the cheapest model that can actually hold a 150K-token context."

Heads up

• GLM 5.1 moved up to the premium pricing band. It's strong but expensive per real token, so the billing now reflects what it actually costs. Pro and Pro Max can still pick it manually; Community can't (Community caps at workhorse-band models).
• Kimi K2.5 is retired. Anything that was pinned to K2.5 routes to K2.6 automatically — quality and price are both better.