Kimi K3 (Max)
An evaluated model label, preserved from the source. Use the evidence below to understand what was measured and what remains unknown.
Source CI: 1482.9 to 1491.9
35,379 votes
Published 2026-10-22
Arena text preference rating (overall, style controlled). Not a percentage or coding pass rate.
Source CI: 1648.9 to 1660.8
20,312 votes
Published 2026-10-23
Arena web development preference rating (overall). Distinct from repository issue resolution or general coding accuracy.
95% CI: 0.0082 to 0.0232
13,076,043 observations · 139,480 sessions
Published 2026-10-23
Arena agent IPS score (overall). Interpret within the source methodology; not interchangeable with text or web development ratings.
Evaluation record
- Source
- arena
- Publisher / namespace
- moonshot
- Prompt mode
- Not supplied
- Source-reported parameters
- Not provided
- Model release date
- Not provided
- Evaluation date
- Not supplied per model
- Source updated
- 22 Oct 2026
- Source revision
- 9761a3ce414bac9ebac5a7742f67f230be809df4
- Reasoning effort / harness
- Not supplied per model in this source
- Confidence intervals
- Shown with each measurement above; overlapping intervals do not establish superiority
- Model license / context limit
- Not verified in this dataset
Price, speed & practical use
Verified pricing is not included in this source. We do not infer current API pricing from historical model names.
Output speed: not measured in this dataset.
Source ratings describe specific evaluations and configurations; they do not establish reliability for every application or agent workflow. Validate candidates on your own workload.
Source: Arena / lmarena-ai leaderboard dataset, CC BY 4.0. Filtered to the latest overall boards and normalized by BuildnScale; evaluations were not rerun. No endorsement implied.
Published evidence. Clear sources. No sponsored rank placement.
How to read these results →