88Z AI, 21 April — Moonshot AI releases open-source model Kimi K2.6, strengthening code generation, long-horizon tasks, and agent swarm performance, with benchmark results comparable to leading closed-source models.
At a glance
- Kimi K2.6 matches or exceeds closed-source models like GPT-5.4 in various benchmarks
- Supports parallel operation of 300 sub-agents and up to 5 days of continuous autonomous running
The update
Moonshot AI has officially open-sourced the Kimi K2.6 model, which features comprehensive upgrades in code writing, long-horizon task execution, and agent swarm collaboration. In multiple authoritative benchmarks such as Humanity’s Last Exam and SWE-Bench Pro, Kimi K2.6 matches or outperforms well-known closed-source models like GPT-5.4 and Claude Opus 4.6, demonstrating strong competitiveness. In terms of practical application, Kimi K2.6 supports continuous coding for 13 hours and can handle modifications to over 4,000 lines of code. Its agent swarm capability is particularly notable, capable of orchestrating 300 sub-agents in parallel to complete up to 4,000 collaborative steps, while supporting sustained autonomous operation for up to five days, significantly improving automation efficiency for complex tasks.
Sources & editorial note
This article is an AI-assisted synthesis by 88Z AI based on the public sources below. Product claims and availability are attributed to their publishers. It does not represent independent testing or first-hand reporting.
- 月之暗面 KimiSource published
Source dates above refer to the original announcements; the date under the headline is this article’s publication on 88Z AI.