Moonshot has released the open weights for Kimi K3.1 Pro, and the model now ranks first among open models on Agent Arena, an evaluation based on real-world, long-horizon agentic sessions.
Open-weight release · Moonshot · Agent Arena
Kimi K3.1 Pro ranks No. 1 among open models on long-horizon agent tasks
Moonshot has released the open weights for Kimi K3.1 Pro, which now leads open models on Agent Arena, an evaluation built from real, multi-step agentic sessions rather than single-turn prompts.
17,500+
Agentic sessions contributed by Arena's global user community
5.1%
Net improvement for the Max variant across those sessions
No. 1
Rank among open models on the Agent Arena leaderboard
Net improvement, drawn to scale
Each block = 1 percentage point of net improvement. The announcement does not disclose the baseline, so this shows the gain magnitude only.
Takeaway: Five full blocks plus a partial sixth. The gain is real in the evaluator's measure, but its size only means something once the baseline and scoring method are published.
Why it matters
Open weights let developers inspect, fine-tune and deploy the model locally.
A top open-model ranking on agentic work is a milestone for the category, where open models have often trailed closed systems.
It adds evidence for testing agents on extended workflows, not short benchmark questions.
What is still unclear
The exact baseline and scoring method behind the 5.1% figure.
Independent third-party replication of the result.
Architecture, parameter count and license terms.
Reading the result
Treat the ranking as the evaluator's result until more methodology and independent testing are available.
Open question 1
Performance across a wider range of tasks
Open question 2
Head-to-head results against leading closed models on the same evaluation
Open question 3
Whether open access speeds up agent development by independent teams
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...