Breaking

Moonshot's Kimi K3.1 Pro Targets November 4 Open-Weight Release, Scores Competitively Against Claude and GPT-6

October 26, 2026 at 01:19 EDT

Foundation Models
Open Source

Moonshot AI has released Kimi K3.1 Pro, a 2.8-trillion-parameter mixture-of-experts model that will become open weights on November 4, marking a significant moment in the race to democratize frontier-class AI capabilities.

Oct 23 • Moonshot AI

Kimi K3.1 Pro Launches with November 4 Open-Weight Deadline

Frontier-class performance from a mixture-of-experts model set to become fully open source in under two weeks.

65.7%

Terminal-Bench 4.0 Score

104B

Active Parameters (per token)

$2/M

Input Token Cost

Context Window (tokens)

Performance vs. Frontier Models

Kimi K3.1 Pro
65.7%

Claude Opus
66.4%

GPT-6 Astra
58.2%

Terminal-Bench 4.0 coding task performance

Key Timeline

October 23: API launch
October 26: Rate limiting; quota reset
November 4: Open weights publication (promised)

Technical Tradeoff: Security Limitations

Moonshot intentionally weakened hacking capabilities before public release. Internal tests: unrestricted model 90.5%, final K3.1 Pro 81% on exploitation tasks—deliberate boundary to avoid advancing exploit development.

Continue reading

The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.