Vercel announced Kimi K3.1 Pro and Kimi K3.1 Pro Highspeed availability in AI Gateway on October 23, 2026. Moonshot lists access through Kimi.ai, Kimi Work, Kimi Code and the Kimi API, with instructions specifying updated mobile apps, Work 3.1.0 or later on Windows or Apple silicon Macs, and /model in Kimi Code. Enterprise access requires an administrator to enable it; consumer plan quotas, regional restrictions and uniform Highspeed access across interfaces remain unclear. Sources: Moonshot, Vercel
Vercel AI Gateway uses model IDs moonshotai/kimi-k3.1-pro and moonshotai/kimi-k3.1-pro-highspeed. It supports the AI SDK, Chat Completions, Responses and Anthropic Messages APIs, plus connected coding agents. Vercel’s October 23 instructions for Claude Code, Codex and Cursor specify installing the latest Vercel CLI, running npx vercel ai-gateway setup and selecting either model. Sources: Vercel
As of October 24, 2026, Kimi API rates per million tokens were $2 for input and $10 for output for Pro, versus $4 and $20 for Highspeed. Vercel says Gateway adds no inference markup or platform fee. Moonshot advertises a million-token context window, but that capacity alone does not establish whether coding agents preserve constraints or make correct changes across a repository. Sources: Moonshot, Vercel
Moonshot reports 65.7% on Terminal-Bench 4.0, 61.2% on its internal Kimi Code Bench 3.0 Long-horizon and a 2–3× speed gain for Highspeed. Its default evaluation settings use maximum reasoning effort, temperature 1.0 and top-p 0.95; coding evaluations use Claude Code harnesses, while Terminal-Bench comparisons mix release reports and leaderboard maxima. LuminaBench says it has not rerun the benchmarks; independent quality and latency remain unestablished, and firm rankings require matched evaluation settings. Sources: Moonshot, LuminaBench
Moonshot promises full model weights by November 4, 2026, with hosted access already available. That promise does not establish downloadable artifacts or release-specific commercial-use, modification and redistribution rights. Developers considering self-hosting should check official artifacts, model cards, checkpoint identities and license terms when released, including whether both variants are covered. Sources: Moonshot