Muzammil Shah (@Muzammil_OX) on X

What it claims

Kimi K3.1 Pro is a Moonshot AI model that scores 55 on the Artificial Analysis Intelligence Index, 11 points above its predecessor, and ranks 4th overall. Beyond the headline score, the post claims stronger performance on agentic, finance and office tasks. It also claims better coding and 3D workflows in three game engines: Unity, Unreal Engine and Godot. The model is described as open-weight. The post's author, Muzammil Shah, argues that if the rankings hold, Kimi K3.1 Pro could become the highest-scoring open-weight model.

How to use it

The model has a 1M-token context window. API pricing is $2 per million input tokens and $10 per million output tokens, which the post describes as lower than before. The prior price is not given. Moonshot plans to release the weights by November 4, 2026. The post gives no licence, parameter count, model identifier or API endpoint.

Limits

This is a third-party post on X, not Moonshot's release notes. The agentic, finance, office, coding and game-engine claims have no benchmark, metric or baseline attached, so they cannot be checked from this page. The predecessor model is not named. The 4th-place ranking is on the Artificial Analysis leaderboard at the time of posting.

Moonshot removed cyber-related data from training to reduce offensive cybersecurity risk. As a result, its internal vulnerability research score fell from 90.5% to 81%. That benchmark is internal, and its setting is not described. Teams that want to use the model for security tooling or vulnerability analysis should expect weaker results there than the predecessor's.

Until the weights ship, the open-weight status is a stated plan, not a release. The licence terms are also unknown.