Skip to content

All Models

Rank #6 / 13

Kimi K3

Kimi K3 is Moonshot's Kimi model (open weights). It is ranked 6 of 13 models with enough coverage on the Visual Intelligence Index.

  • Moonshot
  • Open weights
  • Kimi
  • Video input
  • Structured output
  • Tool calling
  • Streaming
  • Self-hostable
Read the report
71.7%Visual Intelligence Index · Rank #9 of 13

Capability profile

Comparison Summary

Kimi K3 ranks 6 of 13 on the perception index. Immediately ahead is Gemini 3.8 Flash. Immediately behind is Claude Fable 5.1. List price is $0.60 per million input tokens and $2.50 per million output. 7 of 13 listed models cost less to prompt. Latency is 8.2s with a 256K context window and tokens/s. Prices and most scores in this snapshot are placeholders, not sourced figures. Strongest capability in this set is Spatial Reasoning (85%). Weakest is Action / Event Understanding (68%).

Visual Intelligence
81.5%
Availability
Open weights
Input / 1M tokens
$0.60
Output / 1M tokens
$2.50
Latency
8.2s
Context Window
256K
Speed (tokens/s)
Parameters
3T
Time to first token

All benchmark scores

BenchmarkMetricScore95% CISettingSource
Video-MME v2Accuracy, no subtitles
LVBenchTest accuracy
Perception TestOverall accuracy82.4%78.9–85.91 fpsReported
NExT-QAAccuracy (hard split)86.3%85.3–87.21 fpsReported
Q-Bench VideoOverall accuracy
EgoSchemaAccuracy (fullset)78.6%75.0–82.21 fpsReported
UCF101-ADAccuracy39.4%37.9–40.81 fpsReported

Benchmark Scores

Compare reported model scores across each available benchmark or capability index.

13 of 13 models
Filter by model access

Display

Capabilities Index Scores

Compare reported model scores across each available benchmark or capability index.

13 of 13 models
Filter by model access

Display

Usability

4.0/ 5

What it's like to build against this model, scored out of five from the developer-facing capabilities in the dataset.

Reachable1.0

There is an endpoint you can call without hosting anything.

  • First-party API — supported
  • Third-party API — supported
Portable1.0

You can run it yourself, and are not tied to one vendor.

  • Open weights — supported
  • Self-hostable — supported
Multimodal input0.5

It takes the footage directly, rather than frames you extracted.

  • Video ingestion — supported
  • Audio — not supported
Programmable1.0

Output you can parse, and tools it can call on its own.

  • Structured output — supported
  • Tool calling — supported
Operable0.5

Usable interactively and in bulk, not only one call at a time.

  • Streaming — supported
  • Batch API — not supported

Other Moonshot models

Other models from the Moonshot family.

DateModelVisual IntelligenceLatency
Apr 15, 2026Kimi K381.5%8.2s