Leaderboard / Moonshot AI (Kimi)
Kimi-VL-A3B-Thinking-2506
Moonshot AI (Kimi) · Kimi-VL · released 2025-06-21 · data checked 2026-11-02 Needs verification
Kimi-VL-A3B-Thinking-2506 is a 16.4B-parameter Mixture-of-Experts model (2.8B active per token) from Moonshot AI (Kimi) (China), released 2025-06-21 under the MIT license. It doesn't have enough reported benchmarks for an index score yet. It needs about 11 GB of memory at 4-bit with an 8K-token context.
Benchmarks
Each bar shows this model's score. The tick marks the best medium model on that benchmark. Sources: 1 reported by Moonshot AI (Kimi). How we label sources.
Specifications
- Lab
- Moonshot AI (Kimi) (CN · China)
- Architecture
- Mixture-of-Experts
- Total parameters
- 16.4B
- Active parameters
- 2.8B
- Size class
- Medium
- Context window
- 128K tokens
- Input modalities
- text, image, video
- Reasoning
- Reasoning (thinking)
- License
- MIT
- Openness
- Open weights
- Memory to run
- ~11 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 10.1K downloads (30 days) · 387 likes
Compare with
- Kimi-VL-A3B-Thinking-2506 vs LLaDA2.2 mini (16.3B, index 29.7)
- Kimi-VL-A3B-Thinking-2506 vs Apriel 1.6 15B Thinker (14.9B, index 60.9)
- Kimi-VL-A3B-Thinking-2506 vs Phi-4-reasoning-plus (14.7B, index 54.4)
Sources
Something wrong here? Report it.
Run it locally
Uses ggml-org/Kimi-VL-A3B-Thinking-2506-GGUF (Q4_K_M). Swap the tag for another size: BF16, F16, Q4_K_M, Q8_0.
ollama run hf.co/ggml-org/Kimi-VL-A3B-Thinking-2506-GGUF:Q4_K_Mllama-server -hf ggml-org/Kimi-VL-A3B-Thinking-2506-GGUF:Q4_K_Mvllm serve moonshotai/Kimi-VL-A3B-Thinking-2506Memory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 128K context |
|---|---|---|---|
| 16-bit F16 | 34 GB | 35 GB | 38 GB |
| 8-bit Q8_0 | 18 GB | 19 GB | 22 GB |
| 4-bit Q4_K_M | 11 GB | 12 GB | 15 GB |