Kimi Apologizes for K3.1 Pro Launch Issues and Compensates 25% Usage

On 23 October 2026, Moonshot AI released Kimi K3.1 Pro, its most capable open-weight model to date. According to the company's own figures, it scores 65.7% on Terminal-Bench 4.0, just 0.7 points behind Claude Opus 5.5 at 66.4%, at roughly half the price. Full weights were promised by 4 November. DEV Community's coverage framed it as the first open model that is genuinely close to the frontier rather than a cheaper compromise.

Less than 48 hours later, on the early morning of 26 October, Moonshot AI publicly apologized for rate limiting and abnormal quota consumption affecting subscribed users since launch. The company acknowledged fault, stating in effect that the problem was theirs, and rolled out two measures: subscribers with active plans received usage compensation equal to 25% of the monthly total, with quotas reset to 0 where consumption had fallen below that threshold and progress bars adjusted the same day; at the same time, API per-token pricing was reduced.

The official framing did not fully match what users reported. One subscriber on the CNY 199 tier said having the model install a single Skill consumed 70% of a five-hour quota, and that a normal hour of use burned 27% of a rolling five-hour allowance. That tension between reduced per-token pricing and tighter effective limits is where the story now stands.

As of the latest reporting, the situation is paused at compensation plus continued throttling. Full weights are still scheduled for release by 4 November, rate limiting remains in force until then, and it remains to be seen whether the compensation package will hold users over that window.