Model
Model explorer
DeepSeek-V3-0324
OPENDeepSeek · DeepSeek-V3 family · released Mar 24, 2025
March 2025 refresh of V3 with big reasoning/math/coding/front-end gains (AIME 39.6 to 59.4) and improved function calling and JSON output; MIT-licensed.
ReasoningCodingVisionFunction callingTool useAgentic
1365.5
Elo · rank #144
Parameters
671B
Active params
37B (MoE)
Context
128K tokens
Architecture
DeepSeekMoE: 256 routed + 1 shared expert, top-8 routing, 61 layers (3 dense + 58 MoE), Multi-head Latent Attention (MLA)
License
MIT
Languages
—
API price (in/out)
$0.27 / $1.1
Modalities
text
Benchmark results
Bar shows position within the tracked field; marker = field best
best: Claude Opus 4.5 · 89.4%
best: GPT-5.2 · 100.0%
best: Claude Fable 5 · 1505
best: MiniMax-M2 · 75.7%
best: Claude Fable 5 · 1932
best: GPT-5.6 · 94.6%
best: Claude Sonnet 5 · 57.4%
best: MiniMax M3 · 83.0%
best: DeepSeek-V4-Pro (Think Max) · 93.5%
best: GPT-5 · 99.4%
best: Claude Fable 5 · 91.5%
best: DeepSeek-Prover-V2 671B · 7.1%
best: Claude Fable 5 · 95.0%
best: GPT-5.6 · 88.8%
Run it locally
VRAM @ Q4
404 GB
VRAM @ FP16
1342 GB
Fits on (Q4)
Multi-node cluster required
671B MoE far exceeds a single 24GB 4090; typically deployed on multi-GPU servers or via CPU-offload dynamic GGUF quants (~150-250GB range).
Quantizations
GGUF Q4_K_M · AWQ · MLX
DeepSeek-V3 family
Elo progression across releases
API price $0.27/$1.1 · each benchmark row carries its own source badge (see methodology)