Product Launchmultimodal vlmagent swarmopen sourcebenchmarks
Kimi Releases K2.5 Open-Source Visual Agent
9.8
Startup Kimi.ai recently released Kimi K2.5, an open-source multimodal visual agent claiming state-of-the-art scores on multiple agentic and vision benchmarks (HLE 50.2%, BrowseComp 74.9%, VideoMMMU 86.6%). The model supports an Agent Swarm beta with up to 100 sub-agents and 1,500 tool calls, and is offered for local deployment and API access, positioned as a lower-cost alternative to proprietary models.
Key Points
- 1Debuts K2.5 as open-source multimodal VLM scoring top on several agentic and vision benchmarks
- 2Demonstrates high agentic performance and efficiency, with Agent Swarm beta enabling up to 100 sub-agents
- 3Offers practitioners an affordable, locally runnable alternative to proprietary models at roughly $30 monthly
Scoring Rationale
High novelty and usability from open-source, frontier-level performance; tempered by mixed community reactions and noted alignment concerns.
Sources
Public references used for this report.
Practice interview problems based on real data
1,625 SQL & Python problems across 15 industry datasets — the exact type of data you work with.
Try 250 free problems
