DeepSeek R1 vs DeepSeek V3
At a Glance
| Compare |  DeepSeek R1DeepSeek | DeepSeek V3DeepSeek |
|---|---|---|
| Intelligence, Cost, and Efficiency | ||
| IntelligenceHigher is better · MM Intelligence v4.4 | #42 of 570.4 score · 1/3 sources · single-source provisional · missing LiveBench, ARC-AGI-2 · full-core range 0.1–66.8 | UnrankedNot in the 57-model eligible cohort |
| Pricing and Limits | ||
| Input priceFrom · USD / 1M tokens | $0.70Openrouter ↗ · Aug 28, 2026 | $0.2574Openrouter ↗ · Aug 29, 2026 |
| Output priceFrom · USD / 1M tokens | $2.50Openrouter ↗ · Aug 28, 2026 | $0.89Deepinfra ↗ · Oct 19, 2026 |
| Context windowMaximum documented tokens | 164K | 164K |
| Model facts checked | Aug 28, 2026View model evidence → | Aug 28, 2026View model evidence → |
Token prices are the lowest available sourced USD rates; input and output may use different providers. Cost ranking estimates output spend on LiveBench, not a full request bill. Ranking methodology →
Decision Notes
- Workload evidence
- LMArena Text Arena (text-2026-10-15-0ebb411abf00), human-preference rating: DeepSeek R1 1,372.6529, DeepSeek V3 1,332.4603 rating; higher is better. DeepSeek R1 leads this matched row. These results describe the published evaluation workloads, not a universal model winner or an independent Model Markets measurement. View evidence →
- Reference request
- For 1,000 uncached input and 1,000 output tokens on Openrouter (openai; Standard; regions: not specified; quantization: not specified), DeepSeek R1 is USD 0.003200 and DeepSeek V3 is USD 0.001286. Rates observed 2026-08-28 to 2026-08-29. Reference-request estimate v1: (1,000 × input rate + 1,000 × output rate) / 1,000,000. Excludes cache, tool calls, retries, other fees and taxes; this is not measured task cost. View evidence →Published price source 1 ↗ Published price source 2 ↗
- Deployment requirements
- DeepSeek R1 is documented as self-hostable; DeepSeek V3 is documented as self-hostable. Recorded licenses: DeepSeek R1, mit; DeepSeek V3, not reported. If you need local deployment, verify the license and hardware requirements; an unknown field does not establish permission or lack of support. View evidence →
Available Benchmarks
| Benchmark | DeepSeek-R1 | DeepSeek-V3 |
|---|---|---|
| LMArena Text Arenatext-2026-10-15-0ebb411abf00 · arena_rating · leader | 1,372.65100% of row best · rating · deepseek-r1; 95% CI [1367.80631687, 1377.49940563]; votes 18524; rank 181 | 1,332.4697% of row best · rating · deepseek-v3; 95% CI [1327.72698428, 1337.19353550]; votes 21770; rank 224 |
| Overall ResultCounted from the protocol-matched rows above | 1 benchmark winNo overall winner | 0 benchmark winsNo overall winner |
Third-party benchmark Only like-for-like primary-publisher results are shown; raw scores, relative scores, configuration, and token spend remain visible.
Side-by-Side Facts
| Field | DeepSeek-R1 | DeepSeek-V3 |
|---|---|---|
| Developer | DeepSeek | DeepSeek |
| Family | Deepseek R1 | Deepseek V3 |
| Model | DeepSeek-R1 | DeepSeek-V3 |
| Version | DeepSeek-R1 | DeepSeek-V3 |
| Lifecycle | active | active |
| Released | 2025-01-20 | 2024-12-26 |
| Knowledge cutoff | Unknown | Unknown |
| Input modalities | Text | Text |
| Output modalities | Text | Text |
| Context window | 164K | 164K |
| Total parameters | 684.5B | 684.5B |
| Active parameters | 37B | 37B |
| License | mit | Unknown |
| Open weights | Yes | Yes |
| API available | Yes | Yes |
| Self-hostable | Yes | Yes |
| Provider access | Hugging Face (Standard), Openrouter (Standard) | Deepinfra (Standard), Hugging Face (Standard), Openrouter (Standard) |
| Capabilities | chat, generation, reasoning | chat, generation |
DeepSeek R1 Capabilities
DeepSeek V3 Capabilities
Related Comparisons
All text comparisons →| A | Pair | B | Context |
|---|---|---|---|
DeepSeek-R1DeepSeek | vs | cross-developer peerstext | |
DeepSeek-R1DeepSeek | vs |  Phi-4-reasoningMicrosoft | cross-developer peerstext |
DeepSeek-R1DeepSeek | vs | DeepSeek-V4-Pro-BaseDeepSeek | family variantstext |
| vs | DeepSeek-R1DeepSeek | cross-developer peerstext | |
| vs | DeepSeek-R1DeepSeek | cross-developer peerstext | |
DeepSeek-R1DeepSeek | vs | DeepSeek-V3.2DeepSeek | family variantstext |
DeepSeek-R1DeepSeek | vs |  Llama-3.1-405BMeta | cross-developer peerstext |
DeepSeek-R1DeepSeek | vs | cross-developer peerstext | |
DeepSeek-R1DeepSeek | vs |  Kimi-K2-ThinkingMoonshot AI | cross-developer peerstext |
DeepSeek-R1DeepSeek | vs |  GLM-5Z.ai | cross-developer peerstext |
DeepSeek-R1DeepSeek | vs | GLM-5.1Z.ai | cross-developer peerstext |
Primary Evidence
Sources and Freshness
Questions
DeepSeek R1 vs DeepSeek V3 FAQs
Is DeepSeek R1 or DeepSeek V3 better for coding?+
This comparison does not currently contain a protocol-matched coding benchmark for both DeepSeek R1 and DeepSeek V3, so Model Markets cannot name a coding leader from pricing, context size, or capability labels alone.
Which is cheaper, DeepSeek R1 or DeepSeek V3?+
DeepSeek R1 is $0.70 and DeepSeek V3 is $0.2574 per million tokens, so DeepSeek V3 is cheaper on this metric. DeepSeek R1 is $2.50 and DeepSeek V3 is $0.89 per million tokens, so DeepSeek V3 is cheaper on this metric.
Which has a larger context window, DeepSeek R1 or DeepSeek V3?+
Both models have the same sourced context window: 163,840 tokens each.
Which performs better in benchmarks, DeepSeek R1 or DeepSeek V3?+
There is no overall benchmark winner: At least two independently verified, protocol-matched benchmarks are required for an overall winner.
Can DeepSeek R1 or DeepSeek V3 be self-hosted?+
Both models have the same sourced self-hosting status: supported. DeepSeek R1 is open weight; DeepSeek V3 is open weight.
Can DeepSeek R1 and DeepSeek V3 understand images?+
DeepSeek R1 is not documented with image input; DeepSeek V3 is not documented with image input. This reflects supported input modalities, not vision quality.
Which can generate longer answers, DeepSeek R1 or DeepSeek V3?+
DeepSeek R1 has a sourced maximum output of 32,768 tokens; DeepSeek V3's maximum output is not reported. A larger limit cannot be determined.
Do DeepSeek R1 and DeepSeek V3 support reasoning and tool use?+
DeepSeek R1: reasoning. DeepSeek V3: none of these features are definitively sourced. Feature support does not establish relative quality.
Which is available from more inference providers, DeepSeek R1 or DeepSeek V3?+
DeepSeek R1 has 2 sourced provider routes; DeepSeek V3 has 3, so DeepSeek V3 has broader tracked availability.
Which offers better value, DeepSeek R1 or DeepSeek V3?+
There is no universal value winner. Compare the input and output prices above with the matched benchmark result for your workload: cheaper tokens can be offset by different quality, token usage, latency, or provider availability.