Leaderboard / LLM360 / MBZUAI (K2)
K2-V2 Instruct
LLM360 / MBZUAI (K2) · K2 · released 2025-12-03 · data checked 2026-11-01
K2-V2 Instruct is a 70B-parameter dense model from LLM360 / MBZUAI (K2) (Other), released 2025-12-03 under the Apache-2.0 license. It scores 57.9 on the Small LM Index, #4 of 6 large models. It needs about 47 GB of memory at 4-bit with an 8K-token context.
Fully open 70B model (data, code and checkpoints) from MBZUAI.
Benchmarks
Each bar shows this model's score. The tick marks the best large model on that benchmark. Sources: 4 reported by LLM360 / MBZUAI (K2). How we label sources.
Not reported: estimated
What the index model expects from this model's other scores. Not a measurement.
Specifications
- Lab
- LLM360 / MBZUAI (K2) (AE · Other)
- Architecture
- Dense transformer
- Total parameters
- 70B
- Active parameters
- All (dense)
- Size class
- Large
- Context window
- 512K tokens
- Input modalities
- text
- Reasoning
- Hybrid (thinking on/off)
- License
- Apache-2.0
- Openness
- Fully open: weights, training data and code
- Memory to run
- ~47 GB at 4-bit, 8K context (more sizes)
- Hugging Face
- 1.2K downloads (30 days) · 46 likes
Compare with
- K2-V2 Instruct vs Nemotron 3.5 Bolt 74B-A9B (74B, index 73.3)
- K2-V2 Instruct vs DeepSeek-R1-Distill-Llama-70B (70.6B, index 50.1)
- K2-V2 Instruct vs Llama 3.3 70B (70.6B, index 40.8)
- K2-V2 Instruct vs Kolibri-1 (78.1B, index 68.3)
Ranked in
Sources
Something wrong here? Report it.
Run it locally
No GGUF build from a well-known publisher yet.
vllm serve LLM360/K2-V2-InstructMemory needed: weights, context cache and runtime buffers.
| Quantization | 4K context | 32K context | 256K context |
|---|---|---|---|
| 16-bit F16 | 143 GB | 153 GB | 228 GB |
| 8-bit Q8_0 | 78 GB | 87 GB | 162 GB |
| 4-bit Q4_K_M | 46 GB | 55 GB | 130 GB |