DeepSeek-V3 / R1

Rivals GPT-4; R1 = chain-of-thought reasoning

DeepSeek-V3 is a 671B Mixture-of-Experts model that benchmarks close to GPT-4, while R1 adds chain-of-thought reasoning for math and coding tasks. The open weights and MIT license make it one of the most capable freely available models. Best for researchers and developers who need frontier-level reasoning on their own hardware.

FREE TIER: 671B MoE; fully open
COMPANY: DeepSeek
CATEGORY: Open-Source Models › Large Language Models
OPEN SOURCE: Yes
PRIVACY: Self-host for privacy
TAGS: llmreasoning

QUICK START

Try DeepSeek R1 at chat.deepseek.com for free → For local use, download quantized weights via Ollama ('ollama run deepseek-r1') → Self-host for full data privacy

BEST FOR

▸Researchers & developers needing frontier AI on private hardware.
▸Teams needing top-tier math/coding reasoning with full data control.
▸Developers building custom AI apps with powerful, open-source models.

NOT FOR

✕Users without powerful GPUs or technical expertise for self-hosting.
✕Casual users seeking a simple, cloud-hosted, plug-and-play AI solution.

◈ TYPICALLY PAIRED WITH

Tools that appear alongside DeepSeek-V3 / R1 in AIArsenal's curated stacks.

Ollama

Run 100+ LLMs locally from the terminal — completely free

CLI Tools

Continue.dev

Open-source AI code assistant — bring your own model (local or cloud)

Developer Tools

Llama 3.x

Best all-around open LLM — 8B to 405B

Open-Source

Stable Diffusion