DeepSeek

Frontier open and API models from DeepSeek. DeepSeek 的前沿开源与 API 模型。

Official site ↗

Models in the tree

9 featured nodes · 3 other nodes (from DeepSeek ). (22 collapsed variants hidden from the graph)

Featured (graph/timeline)

• DeepSeek Coder V2 Lite Instruct — Jan 2026
• DeepSeek LLM 67B Chat (V1) — Jan 2026
• DeepSeek r1 distill qwen — Series — Jan 2026
• DeepSeek v4 — Series — Jan 2026
• DeepSeek V3.2 — Sparse Attention at Scale — Sep 2025
• DeepSeek R1 — Open-Weight Reasoning Model — Jan 2025
• DeepSeek V3 — 671B Open MoE at $5.5M Training Cost — Dec 2024
• DeepSeek-Coder V2 — Open Coding Frontier — Jun 2024
• DeepSeek-V2 — Open MoE with Multi-Head Latent Attention — May 2024

Series

Benchmark snapshot

Best score within this company’s nodes (not a normalized cross-benchmark ranking).

• GPQA: 87.1% (73.8–87.1%) (from deepseek-v2-5)
• MMLU-Pro: 86.3% (83.6–86.3%) (from deepseek-v2-5)
• Speed · Output tok/s: 47 (from deepseek-r1-distill-llama)
• Coding · SciCode: 44.0% (37.7–44.0%) (from deepseek-v2-5)
• Coding · AA Coding Index: 36.70 (from deepseek-v2-5)
• Agentic · Terminal-Bench Hard: 35.6% (25.0–35.6%) (from deepseek-v2-5)
• Intelligence · AA Intelligence Index: 33.90 (from deepseek-v3-1)
• HLE: 26.1% (8.6–26.1%) (from deepseek-v2-5)

Official catalog (manual)

• deepseek-v4-pro — DeepSeek-V4 Pro (Preview) — in tree
• deepseek-v4-flash — DeepSeek-V4 Flash (Preview) — in tree
• deepseek-reasoner — deepseek-reasoner (legacy alias → V4-Flash thinking) — in tree
• deepseek-chat — deepseek-chat (legacy alias → V4-Flash non-thinking) — in tree
• deepseek-r1-distill-llama-70b — DeepSeek R1 Distill Llama 70B — in tree
• deepseek-v4-flash-high — DeepSeek V4 Flash (Reasoning, High Effort) — in tree
• deepseek-v3-2 — DeepSeek V3.2 (Non-reasoning) — in tree
• deepseek-v3-2-speciale — DeepSeek V3.2 Speciale — in tree
• deepseek-r1 — DeepSeek R1 0528 (May '25) — in tree
• deepseek-v3-2-reasoning — DeepSeek V3.2 (Reasoning) — in tree
• deepseek-v4-pro-high — DeepSeek V4 Pro (Reasoning, High Effort) — in tree
• deepseek-r1-distill-qwen-32b — DeepSeek R1 Distill Qwen 32B — in tree
• deepseek-v3 — DeepSeek V3 (Dec '24) — in tree
• deepseek-r1-distill-qwen-14b — DeepSeek R1 Distill Qwen 14B — in tree
• deepseek-v2-5 — DeepSeek-V2.5 (Dec '24) — in tree
• deepseek-coder-v2 — DeepSeek-Coder-V2 — in tree
• deepseek-r1-distill-llama-8b — DeepSeek R1 Distill Llama 8B — in tree
• deepseek-llm-67b-chat — DeepSeek LLM 67B Chat (V1) — in tree
• deepseek-r1-distill-qwen-1-5b — DeepSeek R1 Distill Qwen 1.5B — in tree
• deepseek-v3-0324 — DeepSeek V3 0324 — in tree
• deepseek-v3-1-reasoning — DeepSeek V3.1 (Reasoning) — in tree
• deepseek-r1-0120 — DeepSeek R1 (Jan '25) — in tree
• deepseek-v3-1-terminus — DeepSeek V3.1 Terminus (Non-reasoning) — in tree
• deepseek-v3-1 — DeepSeek V3.1 (Non-reasoning) — in tree
• deepseek-v3-1-terminus-reasoning — DeepSeek V3.1 Terminus (Reasoning) — in tree
• deepseek-v2 — DeepSeek-V2-Chat — in tree
• deepseek-v3-2-reasoning-0925 — DeepSeek V3.2 Exp (Reasoning) — in tree
• deepseek-coder-v2-lite — DeepSeek Coder V2 Lite Instruct — in tree
• deepseek-v3-2-0925 — DeepSeek V3.2 Exp (Non-reasoning) — in tree
• deepseek-v2-5-sep-2024 — DeepSeek-V2.5 — in tree
• deepseek-r1-qwen3-8b — DeepSeek R1 0528 Qwen3 8B — in tree

Sources

• https://www.deepseek.com/
• https://api-docs.deepseek.com/updates

Notes

This page is auto-assembled from nodes in the tree whose org matches one of the orgs above.