Maker of DeepSeek V4-Pro, the top-ranked open-source coding model. DeepSeek’s coding models on Kilo Code include DeepSeek V3.1 Terminus, DeepSeek V4 Flash, and DeepSeek V4 Pro. Use them across VS Code, JetBrains IDEs, Cursor, Windsurf, Trae, and the Kilo CLI — with pay-as-you-go pricing and no markup over the underlying provider rates.
Sorted by coding-index where published. Click any model for the full review with benchmarks, real-world Kilo usage, and provider-specific pricing.
DeepSeek
DeepSeek-V3.1 Terminus is an update to [DeepSeek V3.1](/deepseek/deepseek-chat-v3.1) that maintains the model's original capabilities while addressing issues reported by users, including language consistency and agent capabilities, further optimizing the model's...
DeepSeek
DeepSeek V4 Flash is an efficiency-optimized Mixture-of-Experts model from DeepSeek with 284B total parameters and 13B activated parameters, supporting a 1M-token context window. It is designed for fast inference and...
DeepSeek
DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window. It is designed for advanced reasoning, coding,...
DeepSeek is the Chinese AI lab whose V4 series leads every published open-source coding benchmark in 2026. DeepSeek V4-Pro is a 1.6T-parameter MoE model with a true 1M-token context window and the highest published scores on LiveCodeBench (93.5) and Codeforces (3206) of any model, open or closed. All weights ship under MIT.
Pay-as-you-go, no markup over the underlying provider rates. Cheapest first.
| Model | Input / 1M | Output / 1M | Context | Coding index |
|---|---|---|---|---|
| DeepSeek: DeepSeek V4 Flash | $0.200 | $0.400 | 1049K | — |
| DeepSeek: DeepSeek V3.1 Terminus | $0.270 | $0.950 | 164K | — |
| DeepSeek: DeepSeek V4 Pro | $2.40 | $4.80 | 1049K | — |
Three ways: hosted in Kilo, locally on your hardware, or through your own provider keys.
The fastest path: install Kilo Code, sign in, pick deepseek from the model picker. No API keys, no markup. Works in VS Code, JetBrains, Cursor, Windsurf, Trae, and the Kilo CLI.
See live model leaderboard →Already have an account with DeepSeek, OpenRouter, AWS Bedrock, Google Vertex, Together AI, or another compatible provider? Plug your key into Kilo Code and keep your existing billing relationship.
BYOK setup guide →Download DeepSeek open weights from Hugging Face and serve them with Ollama, LM Studio, vLLM, or SGLang. Connect Kilo Code to your local OpenAI-compatible endpoint and keep all prompts on hardware you control.
Local setup guide →See coding-model lineups from DeepSeek’s closest competitors.
DeepSeek’s flagship is among the strongest coding models on the Kilo Code leaderboard, ranked by Code, Plan, Ask, Debug, and Review usage.
VS Code, Cursor, Windsurf, Trae, JetBrains IDEs (IntelliJ, PyCharm, WebStorm, GoLand, RubyMine, Android Studio), and the Kilo CLI / terminal.
Switch between DeepSeek and 500+ other models with one click. Pay only for what you use, at the underlying provider rate.
DeepSeek V4-Pro is the strongest DeepSeek coding model on Kilo Code: the best LiveCodeBench and Codeforces scores of any model evaluated, plus a true 1M-token context window. V4-Flash is the fast cheap variant; V3.1 Terminus is the previous-generation workhorse.
Every DeepSeek model works in all Kilo-supported surfaces: VS Code, Cursor, Windsurf, Trae, JetBrains IDEs, and the Kilo CLI. DeepSeek additionally runs on local or self-hosted infrastructure through any OpenAI-compatible runtime.
DeepSeek pricing on Kilo Code is among the lowest of any frontier-class model. V4-Pro is $1.60 per million input and $3.38 per million output. V4-Flash is $0.14 / $0.28. V3.1 Terminus is $0.27 / $0.95. All run with no markup over the underlying provider rates.
Yes. Every DeepSeek model is published as open weights under the MIT license. You can download V4-Pro, V4-Flash, and V3.1 Terminus from Hugging Face and self-host with vLLM, SGLang, or other OpenAI-compatible runtimes — or use them hosted in Kilo Code with no setup.
V4-Pro is the 1.6T MoE flagship with 1M context, FP4/FP8 mixed precision, and #1 LiveCodeBench / Codeforces scores. V4-Flash is a smaller faster sibling for everyday tasks. V3.1 Terminus is the previous V3-series flagship, still strong for general coding work at lower cost.
Every DeepSeek model works in all Kilo-supported surfaces: VS Code, Cursor, Windsurf, Trae, JetBrains IDEs, and the Kilo CLI. DeepSeek additionally runs on local or self-hosted infrastructure through any OpenAI-compatible runtime.
Install Kilo Code and get instant access to DeepSeek: DeepSeek V3.1 Terminus and 2 other DeepSeek models, plus 500+ frontier and open-source options. Free to start, no credit card required.