DeepSeek-V3 vs Kimi-K2-Thinking-NVFP4
Specs, VRAM requirements and download trends — updated 25 October 2026.
Specification comparison
Differences are highlighted; identical values are muted.
| Specification | DeepSeek-V3 | Kimi-K2-Thinking-NVFP4 |
|---|---|---|
| Deployment | Open weights · also via API | Open weights |
| Parameters | 684.5 B | 519.1 B |
| Architecture | DeepseekV3ForCausalLM | DeepseekV3ForCausalLM |
| Context length | 163,840 | 262,144 |
| Licence | — | other |
| Commercial use | Unknown | Unknown |
| Languages | — | — |
| Downloads 30d | 1,420,996 | 228,965 |
| Downloads all time | 22.6 M | 652 K |
| Quantizations on the Hub | SAFETENSORS | SAFETENSORS |
| Gated | No | No |
| First seen on the Hub | 2024-12-25 | 2025-12-02 |
| API input price / 1M tokens | $0.26 | not on OpenRouter |
| API output price / 1M tokens | $1.03 | not on OpenRouter |
| Est. cost per 1M input + 1M output tokens | $1.29 | not on OpenRouter |
| Cached input / 1M tokens | — | not on OpenRouter |
| AA Intelligence Index | — | not on OpenRouter |
| AA Coding Index | — | not on OpenRouter |
| AA Agentic Index | — | not on OpenRouter |
| Max output tokens | 16,000 | not on OpenRouter |
| API providers | 2 | not on OpenRouter |
| Tool calling | Yes | not on OpenRouter |
| Reasoning mode | No | not on OpenRouter |
API cost
Estimated cost of 1M input + 1M output tokens via OpenRouter
DeepSeek-V3
$1.29
Download trend
Daily snapshots, last 90 days (28 Jul – 25 Oct)
1.5 M171 K
DeepSeek-V3
Kimi-K2-Thinking-NVFP4
VRAM side by side
On RTX 4090 · 24 GB · 8K context unless noted
DeepSeek-V3 · fp16 @ 160K ctx
2,811.5 GB / 24 GB
Kimi-K2-Thinking-NVFP4 · fp16 @ 256K ctx
3,145.7 GB / 24 GB