DeepSeek R1
Open reasoning model trained for hard multi-step math, science, and code.
DeepSeek’s reasoning-focused model trained with reinforcement learning for multi-step math, science, and coding problem solving.
Why DeepSeek R1 matters
DeepSeek R1 showed that open (or open-weight) reasoning traces can rival closed thinking models on hard problems—at attractive cost. It reset expectations for what open reasoning stacks can deliver.
Tool calling · Thinking · Coding
Last reviewed: 31 July 2026
When to choose DeepSeek R1
Decision guidance for architects—not a feature list.
Best for
- Hard math and science questions
- Competitive-programming style tasks
- Chain-of-thought heavy agents
- Open reasoning baselines
Avoid if
- You need strong multimodal (vision/audio) in one model
- Low latency without long thinking traces is mandatory
Strengths
DataAIHub qualitative scores from 1 to 5, for architects. Not official benchmark results or a public ranking.
- Hard-problem reasoning★★★★★
- Open weights / research★★★★★
- Cost for reasoning★★★★★
- Multimodal★☆☆☆☆
- Latency when thinking★★☆☆☆
Ecosystem
Built by
- DeepSeek
R1 is DeepSeek’s reasoning-focused open-weight flagship.
Competes with
Works with
- vLLM
Common serving path for DeepSeek open weights.
Recommended for
- Generative AI
Landmark open reasoning model for modern generative AI stacks.
Often paired with
- DeepSeek V4
V4 is the current general/agent API generation; R1 remains the dedicated open reasoner.
- DeepSeek V3
V3 often handles general chat; R1 handles hard reasoning steps.
How DeepSeek R1 evolved
Key moments in chronological order.
- Benchmark
Reasoning benchmark wave
GPQA / LiveBench-style comparisons put R1 on the open-reasoning map.
- Research
DeepSeek-R1 technical report
RL-trained reasoning model with public traces reshapes open-model expectations.
- Open source
Open weights release
Weights and distillations enable self-hosted reasoning baselines worldwide.
- API
DeepSeek API availability
Hosted API makes R1 accessible without local GPU fleets.
- Model
V3 generalist context
Strong MoE generalist precedes R1 and pairs with it in many stacks.
Overview
Capabilities
- Vision: No
- Audio input: No
- Tool calling: Yes
- Thinking: Yes
- MCP: No
- Coding: Yes
- Structured output: Yes
Technical specifications
- Provider
- DeepSeek
- License
- Model License (open weights)
- Context window
- 128K
- Parameters
- Reasoning MoE family
- Architecture
- Reasoning / RL-trained
- Release
- 2025
- Modalities
- Text
- Vision
- No
- Audio input
- No
- Tool calling
- Yes
- Thinking
- Yes
- MCP
- No
- Open weights
- Yes
- API
- Yes
- Pricing (input)
- DeepSeek API
- Pricing (output)
- DeepSeek API
Supported modalities
Text
Context window
128K (128,000 tokens)
Pricing
Availability
Use cases
- Hard math and science questions
- Competitive programming style tasks
- Chain-of-thought heavy research agents
- Open reasoning baselines
Strengths
- Strong reasoning traces and hard-problem performance
- Open weights for research and self-hosting
- Attractive cost for reasoning workloads
Limitations
- Longer outputs / higher latency when thinking
- Less multimodal than proprietary frontier models
Related guides
Related benchmarks
Related research
- DeepSeek-R1 technical report
Original Paper
- Top Open-Source & Open-Weight AI Models 2026
Evaluation
Related GitHub
Related tools
Related rankings
Companies
Explore more models
- DeepSeek V3DeepSeekDeepSeek’s MoE general model — strong open-weight performance on coding and knowledge tasks with competitive API pricing.
- DeepSeek V4DeepSeekDeepSeek’s V4 generation — live Flash is V4.1-Flash-1017 (API id deepseek-flash, upgraded 2026-10-17): Causal Encoder–Decoder MoE, 552B backbone (8B active prefill / 16B decode), native image+text, and 1M context. API id and prices are unchanged from the 2026-09-10 Flash list. MIT weights on Hugging Face. Legacy deepseek-v4-flash and deepseek-v4-flash-vision-exp aliases temporarily route here. deepseek-v4-pro still serves V4-Pro-0813 after 2026-09-14 at unchanged Pro rates (DeepSeek withdrew the planned Flash reroute).
- Beam 1.0ReflectionReflection’s Beam 1.0 — Apache-2.0 text-only sparse MoE (501B total, 23B active, 1M context) for agentic coding and reasoning. Weights shipped 2026-10-16 with day-0 support in Hugging Face Transformers, vLLM, and SGLang. This is not the beam.cloud GPU tool.
- Sarvam 105BSarvamSarvam’s flagship open-weight reasoning model. A Mixture-of-Experts transformer with 10.3B active parameters, trained from scratch in India, and used to power the Indus assistant.
- Sarvam 30BSarvamSarvam’s efficient open-weight reasoning MoE. About 2.4B active parameters, built for real-time conversational agents on the Samvaad platform and for local or GPU-constrained deployment.
- Kimi K3Moonshot AIMoonshot’s Kimi K3 — 2.8T MoE (104B active) open-weight multimodal agentic model with 1M context, native vision, and strong long-horizon coding. Weights on Hugging Face under the Kimi K3 License.