Moonshot ships Kimi K3.1 Pro via API, promises open weights by Nov 4

Moonshot's Kimi K3.1 Pro, a 1M-context model aimed at coding agents, is live on its API and apps, with full weights promised by November 4.

Moonshot released Kimi K3.1 Pro on October 23–24, 2026, a post-trained version of Kimi K3 with a 1-million-token context window. It is live now on Kimi.ai, Kimi Work, Kimi Code and the Kimi API, and Moonshot says the full weights will be released by Nov 4. For anyone running coding agents, the draw is terminal-task performance: Moonshot reports 65.7% on Terminal-Bench 4.0, against its listed 66.4% for Claude Opus 5.5. Moonshot ran its model with the Claude Code harness and took rivals' scores from their own reports or leaderboards. Artificial Analysis measured 67% on the same benchmark in its own runs and ranks the model 4th on its Intelligence Index with a score of 55.

API pricing is $2 per million input tokens and $10 per million output, a third below K3. The catch is token use. Artificial Analysis calls the model "notably slow" and "very verbose": it generated 150M tokens on the index against a median of 81M and cost $2.11 per task. OrcaRouter notes that GPT-6.1 Sol has the same list price and scores 52, but costs $0.72 per task on Artificial Analysis's numbers.

What's unclear: the weights aren't out yet, and Yotta Labs reports the license hasn't been announced. At 2.8 trillion total parameters, Notebookcheck points out it won't run on a regular PC, so most people will use Moonshot's servers. Moonshot says zero data retention is available for API customers. Until the weights and license land, treat this as a hosted model and budget for its output tokens, not the rate card.

Sources