DeepSeek V4
LLM Provider / Model · Low-cost coding LLM with self-host option; strong English + Chinese coding capabilities
At a glance
| Input price | $0.27 |
| Output price | $1.10 |
| Cache price | $0.07 |
| Price tier | Budget |
| Context window | 256K |
| Max output | 16K |
| Context tier | 128K-500K |
| Speed tier | Medium |
| Latency | medium |
| Knowledge cutoff | Oct 2025 |
| Modality | Text |
| Model ID | deepseek-v4 |
| Provider | DeepSeek |
| HumanEval | 92% |
| MMLU | 86% |
| SWE-Bench | 63% |
| Benchmark | Beats most open-weight models at coding and math |
| Released | Nov 2025 |
| Hosting | Open/Self-host |
| Capabilities | Tool use, MoE architecture, Function calling, Streaming, Prompt caching, Self-host friendly |
What DeepSeek V4 does
Tool use, MoE architecture, Function calling, Streaming, Prompt caching, Self-host friendly
Best for
Low-cost coding LLM with self-host option; strong English + Chinese coding capabilities
Works well with
Integration
Agent / Orchestration
Conflicts & caveats
- Google Antigravity is built around Gemini 3 Pro; Claude and recent GPT-5 models also work, but DeepSeek V4 is not in the supported model list.
- Kiro's spec mode is tuned for Claude models — DeepSeek V4 may work but spec-driven flows are best with Claude.
- OpenAI Codex (VS Code) is tied to OpenAI's GPT-5.1 Codex family — DeepSeek V4 is not selectable inside this integration.
- Gemini CLI is locked to Google Gemini models — your LLM is DeepSeek V4.
- Privacy conflict: Self-hosted DeepSeek V4 sends code to cloud @codebase (Cursor). Use local context (Continue indexing, ChromaDB, LanceDB, pgvector, Vespa self-hosted) for true privacy.
- Claude Code only works with Claude models — your LLM is DeepSeek V4.
- Devin uses Cognition's proprietary model internally — your selected DeepSeek V4 will be ignored.
- OpenAI Codex (Cloud) runs GPT-5.1 Codex internally — your selected DeepSeek V4 will be ignored.
- Jules runs Gemini models internally — your selected DeepSeek V4 will be ignored.
- The Claude Code SDK is tuned for Claude models — DeepSeek V4 may work via the model adapter but loses Claude-specific features (thinking, prompt caching, sub-agents).
- Lovable uses a fixed internal model — your selected DeepSeek V4 will be ignored.
- Replit Agent 3 uses its own model selection internally — your selected DeepSeek V4 will be ignored.
Build a full stack around DeepSeek V4 — Flowpicker shows compatibility warnings before you commit.
Open the stack planner →