Moonshot Kimi K3.1 Pro Tops Benchmarks Amid Influencer Allegations

"Please DO NOT put on the paid partnership label."

That instruction, lifted from screenshots an X user going by Elshayib described as a brief for a Kimi promotional campaign, is dragging on an otherwise historic Chinese model release. In VentureBeat's coverage, Carl Franzen writes that Elshayib "posted screenshots he described as instructions for a Kimi promotional campaign, saying he declined to participate on those terms." The document has not been independently authenticated, and no confirmed Moonshot response had surfaced as of October 25.

The release itself was built to tell its own story. Moonshot unveiled Kimi K3.1 Pro on Friday, a 3-trillion parameter post-trained version of July's K3. On Artificial Analysis's Intelligence Index it scored 55, ahead of Claude Opus 5.5, GPT-6 Astra, DeepSeek V4.1-Flash-1017 and Meta's Muse Wonder. The weights go open under license on November 4, 2026. If the scores hold, Franzen writes, Kimi "will almost certainly take the crown as the top open weights AI model in the world."

The gains over K3 are large. Terminal-Bench 4.0 moved from 13% to 67%, CyberGym-E2E-AA from 58% to 95%, AutomationBench-AA from 58% to 73%. On price, Kimi lists $2 per million input tokens and $10 per million output: roughly half Opus 5.5, but "Kimi charges about 6.7 times as much for uncached input and 8.3 times as much for output" against DeepSeek at peak rates.

Franzen keeps the two threads apart. "The allegations do not, by themselves, invalidate independent benchmark results," he writes. The dispute is over how the launch was promoted, not the Artificial Analysis number itself. It lands in a crowded stretch for the open-source beat; our tracker has logged 357 open-source stories in the last 90 days.