Moonshot's Kimi team has released the open weights for Kimi K3.1 Pro, which Arena says now ranks first among open models on its Agent Arena leaderboard with a +5.3% net improvement.
Weekly leaderboard update · Arena
Kimi K3.1 Pro opens its weights and ranks first among open models on Agent Arena
Moonshot's Kimi team released the open weights for Kimi K3.1 Pro, which Arena places at the top of its open-model Agent Arena leaderboard. The same roundup named Tencent's Hy4 the top-ranked Chinese model.
#1
Kimi K3.1 Pro, first among open models on Agent Arena
+5.3%
Reported net improvement (baseline not specified)
Hy4
Top-ranked Chinese model in the same roundup (Tencent Hunyuan)
The reported +5.3% gain, drawn to scale
1 block = 1 percentage point. The baseline is not disclosed, so this shows the size of the reported figure only, not a comparison against any other model.
+5.3%
Five full blocks plus a partial block (0.3). Baseline and methodology not published.
What open weights make possible
STEP 1
Download
Get the model weights directly
→
STEP 2
Run
Self-host instead of using a hosted API
→
STEP 3
Fine-tune
Adapt the model to your own tasks
Why it matters
Open-weight models can be deployed by anyone, not only through a vendor API.
A top open-model rank on an agent-style benchmark signals strong performance on that task type.
Enterprises weighing self-hosted alternatives to proprietary systems are watching closely.
Caveats
Independent reception and real-world testing were not part of the review.
Benchmark rankings reflect specific task mixes and can shift as new models are submitted.
Performance outside the benchmark remains untested.
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...