Deepseek-V4-Pro

Model IntroductionText

DeepSeek-V4: Towards Highly Efficient Million-Token Context Intelligence

We present a preview version of DeepSeek-V4 series, including two strong Mixture-of-Experts (MoE) language models — DeepSeek-V4-Pro with 1.6T parameters (49B activated) and DeepSeek-V4-Flash with 284B parameters (13B activated) — both supporting a context length of one million tokens.

DeepSeek-V4 series incorporate several key upgrades in architecture and optimization:

  1. Hybrid Attention Architecture: We design a hybrid attention mechanism combining Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA) to dramatically improve long-context efficiency. In the 1M-token context setting, DeepSeek-V4-Pro requires only 27% of single-token inference FLOPs and 10% of KV cache compared with DeepSeek-V3.2.
  2. Manifold-Constrained Hyper-Connections (mHC): We incorporate mHC to strengthen conventional residual connections, enhancing stability of signal propagation across layers while preserving model expressivity.
  3. Muon Optimizer: We employ the Muon optimizer for faster convergence and greater training stability.

We pre-train both models on more than 32T diverse and high-quality tokens, followed by a comprehensive post-training pipeline. The post-training features a two-stage paradigm: independent cultivation of domain-specific experts (through SFT and RL with GRPO), followed by unified model consolidation via on-policy distillation, integrating distinct proficiencies across diverse domains into a single model.

Benchmarks

Benchmark (Area)Score*
HLE (Reasoning)37.7 (Eval)
HLE (w/ Tools) (Reasoning)48.2 (Eval)
CritPt (Reasoning)12.9 (Eval)
AIME 2026 (Reasoning)94.6 (Eval)
HMMT Nov. 2025 (Reasoning)94.4 (Eval)
HMMT Feb. 2026 (Reasoning)95.2 (Eval)
IMOAnswerBench (Reasoning)89.8 (Eval)
GPQA-Diamond (Reasoning)90.1 (Eval)
SWE-bench Pro (Coding)55.4 (Eval)
NL2Repo (Coding)35.5 (Eval)
DeepSWE (Coding)8 (Eval)
ProgramBench (Coding)47.8 (Eval)
Terminal Bench 2.1 (Terminus-2) (Coding)64 (Eval)
Terminal Bench 2.1 (Best Reported Harness) (Coding)- (Eval)
FrontierSWE (Dominance) (Coding)29.0 (Eval)
PostTrainBench (Coding)- (Eval)
SWE-Marathon (Coding)- (Eval)
MCP-Atlas (Public Set) (Agentic)73.6 (Eval)
Tool-Decathlon (Agentic)52.8 (Eval)

API request example

Endpoint /v1/chat/completions. Replace <API_KEY> with a key from the console.

Standard OpenAI Chat Completions. Swap `model` for the target slug.

curl https://api.ravenrouter.ai/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer <API_KEY>" \
  -d '{
  "model": "Deepseek-V4-Pro",
  "messages": [
    {
      "role": "user",
      "content": "Hello! Introduce yourself in one sentence."
    }
  ],
  "max_tokens": 64
}'

More models: API Reference → Examples by model

Deepseek-V4-Pro API — $1.35/1M input, $2.70/1M output | RavenRouter AI