Kimi-K2-Instruct-0905
Kimi K2-Instruct-0905 is the latest, most capable version of Kimi K2. It is a state-of-the-art mixture-of-experts (MoE) language model, featuring 32 billion activated parameters and a total of 1 trillion parameters.
Params
1,026.5 B
Context
262,144
Downloads 30d
32 K
Likes
790
Download history
tracking started — chart appears after 7 days of snapshots (5 recorded)32 K downloads in the last 30 days
Can you run it?
Estimated VRAM at 8K context unless noted. Pick your hardware to see the verdict per quantization.
| File | Quant | Size | Est. VRAM | Verdict on RTX 4090 · 24 GB |
|---|---|---|---|---|
| model.safetensors | f8_e4m3 | 1,029.2 GB | 1,286.6 GB | ❌ Won’t fit |
| model.safetensors (bf16, full) | bf16 + 256K ctx | 1,029.2 GB | 6,059.7 GB | ❌ Won’t fit |
Estimate: file size × 1.1 + KV cache at 8K + 0.5 GB overhead. Not a benchmark — how we calculate this.
Run it
copy-paste, exact tags checked against the Hub$ curl -s https://aimodelscomparison.com/api/v1/models/kimi-k2-instruct-0905
{
"hf_id": "moonshotai/Kimi-K2-Instruct-0905",
"params_b": 1026.47,
"context_length": 262144,
"license": { "id": "other", "commercial": "unknown" },
"downloads_30d": 31918,
"vram_estimates": [
{ "quant": "f8_e4m3", "gb": 1286.6 }
],
"updated_at": "2026-10-19T06:00:40Z"
}
Use via API
via OpenRouter · USD per 1M tokens · synced 2026-10-24$0.60
input / 1M
$2.50
output / 1M
—
cached input / 1M
$3.10
1M in + 1M out
- Context window
- 262,144 tokens
- Max output
- 98,304 tokens
- Providers
- 1
- Cheapest provider
- Novita · $0.60 / $2.50
- Best throughput
- —
- AA Intelligence / Coding / Agentic
- 15.3 / — / —
✓ Tool calling
✗ Reasoning
✓ Structured outputs
✗ Image input
✗ File input
✗ Audio input
✗ Prompt caching
Specifications
- Architecture
- DeepseekV3ForCausalLM
- Parameters
- 1,026.5 B
- Tensor type
- F8_E4M3
- Context length
- 262,144
- Vocabulary
- 163,840
- Layers / heads
- 61 / 64
- Licence
- other
- First seen on the Hub
- 2025-09-03
- Training datasets
- undisclosed
- Added to our catalog
- 2026-10-20
Compare with any text-generation model