DeepSeek
DeepSeek ranks #7 of 91 providers on BenchLeader by its best model, with 22 ranked models. Its best model is DeepSeek V4.1 Flash 1017 at 64.0 on the BenchLeader Index, strongest in composite. Prices run from $0.068 to $2.50 per million tokens blended; DeepSeek-R1-0528-Qwen3-8B is the cheapest ranked option. The fastest is DeepSeek V4.1 Flash 1017 at 195 tokens per second. 24 of its 35 models ship open weights. It has shipped 14 dated releases, roughly one every 37 days; the last was 17 Oct 2026, so the next would be due between 8 Nov 2026 and 7 Dec 2026 on that pattern.
Open-weights lab known for cost-efficient training (V3, R1, V4) and aggressive API pricing.
- Provider rank
- #7
- by best model
- Best index
- 64.0
- DeepSeek V4.1 Flash 1017 (max)
- Ranked models
- 22
- 24 open weights
- Cheapest ranked
- $0.068/M
- DeepSeek-R1-0528-Qwen3-8B
Release timeline
Each model’s best configuration by release date. The top eight are labelled; hover any point for its name.
- DeepSeek22
Next release
Over the last two years DeepSeek shipped 14 distinct models, one every 37 days on median. The latest arrived on 17 Oct 2026.
If the cadence holds, the next one lands between 8 Nov 2026 and 7 Dec 2026.
An estimate from past release dates only. It knows nothing about roadmaps or announcements.
Fastest ranked model: DeepSeek V4.1 Flash 1017 at 195 tok/s.
Head to head
DeepSeek vs Anthropic
DeepSeek vs Google
DeepSeek vs Meta
DeepSeek vs OpenAI
DeepSeek vs Moonshot AI
DeepSeek vs SpaceXAI
DeepSeek vs Xiaomi
DeepSeek vs Zhipu AI
DeepSeek vs Alibaba
DeepSeek vs MiniMax
DeepSeek vs StepFun
DeepSeek vs Sapiens AI
DeepSeek vs ByteDance
DeepSeek vs Motif Technologies
DeepSeek vs Mistral AI
DeepSeek vs MBZUAI
DeepSeek vs Meituan
DeepSeek vs Multiverse Computing
DeepSeek vs NVIDIADeepSeek models compared
Models
Every DeepSeek model with a result, one row per model; switch to all configurations to see each reasoning effort.
| # | ||||||||
|---|---|---|---|---|---|---|---|---|
| 1 | DeepSeek V4.1 Flash 1017maxopen ↗ | 64.0±4 | $0.525 | 195 | 11 s | 1.0M | 2026-10-17 | 2026-10-23 |
| 2 | DeepSeek V4.1 Flashmaxsuperseded ↗open ↗ | 61.2±4 | $0.525 | 181 | 12 s | 1.0M | 2026-09-10 | 2026-10-23 |
| 3 | DeepSeek V4 Pro(Aug 2026)maxopen ↗ | 60.8±3 | $1.98 | 97 | 22 s | 1.0M | 2026-08-13 | 2026-10-20 |
| 4 | DeepSeek V4 Pro 0424maxopen | 59.6±5 | $0.548 | 102 | 45 s | 1M | 2026-04-24 | 2026-04-24 · stale |
| 5 | DeepSeek V4 Flash(Jul 2026)maxopen ↗ | 59.4±4 | $0.175 | 149 | 15 s | 1.0M | 2026-07-31 | 2026-07-31 |
| 6 | Deepseek v4 Flash Visionmax | 58.2±5 | $0.660 | 190 | 12 s | 1.0M | 2026-08-21 | 2026-08-21 |
| 7 | DeepSeek V4 Flash 0420maxopen | 57.0±6 | $0.140 | 47 | 120 s | 1M | 2026-04-24 | 2026-04-24 · stale |
| 8 | DeepSeek V3.2open ↗ | 56.6±6 | $0.315 | 17 | 2.06 s | 131k | 2025-12-01 | 2026-10-23 |
| 9 | DeepSeek V3 0324open ↗ | 53.5±1 | – | – | – | – | 2025-03-24 | 2026-10-22 |
| 10 | DeepSeek V3.1 Terminusthinkingopen ↗ | 53.2±4 | $1.92 | 20 | 1.89 s | 164k | 2025-09-22 | 2026-10-22 |
| 11 | DeepSeek V3.2 Specialeopen ↗ | 53.0±5 | $0.855 | – | – | 128k | 2025-12-01 | 2025-12-01 · stale |
| 12 | DeepSeek R1 0528open ↗ | 49.5±4 | $2.08 | 33 | 62 s | 164k | 2025-05-28 | 2026-10-22 |
| 13 | DeepSeek V3.1thinkingopen ↗ | 48.1±4 | $0.855 | 120 | 19 s | 128k | 2025-08-21 | 2025-08-21 · stale |
| 14 | DeepSeek R1(Jan 2025)superseded ↗open ↗ | 46.8±4 | $2.50 | 144 | 17 s | 66k | 2025-01-20 | 2026-10-22 |
| 15 | Deepseek Coder v2open ↗ | 44.9±3 | – | – | – | 128k | 2024-06-17 | 2026-10-22 |
| 16 | DeepSeek V3(Dec 2024)superseded ↗open ↗ | 44.0±3 | $0.463 | 54 | 1.07 s | 66k | 2024-12-26 | 2026-10-22 |
| 17 | DeepSeek-R1-Distill-Qwen-32Bopen ↗ | 42.5±8 | $0.431 | – | – | 128k | 2025-01-20 | 2025-01-20 · stale |
| 18 | R1 Distill Llama 70Bopen ↗ | 41.5±6 | $0.800 | – | – | 128k | 2025-01-20 | 2025-01-20 · stale |
| 19 | DeepSeek-R1-Distill-Qwen-14Bopen ↗ | 41.2±10 | $0.216 | – | – | 128k | 2025-01-20 | 2025-01-20 · stale |
| 20 | DeepSeek-R1-0528-Qwen3-8Bopen ↗ | 37.6±7 | $0.068 | – | – | 128k | 2025-05-29 | 2025-05-29 · stale |
| 21 | DeepSeek-R1-Distill-Qwen-1.5Bopen ↗ | 36.1±5 | – | – | – | 128k | 2025-01-20 | 2025-01-20 · stale |
| 22 | DeepSeek LLM 67Bopen | 35.4±3 | – | – | – | 4k | 2023-11-29 | 2026-10-22 |
| 23 | Deepseek R1 Plus Claude 3.5 Sonnet | – | – | – | – | – | – | 2025-01-23 · stale |
| 24 | Deepseek v2open ↗ | – | – | – | – | 128k | 2024-05-07 | 2024-05-06 · stale |
| 25 | DeepSeek V3.2 Reasoneropen | – | – | – | – | – | 2025-12-01 | 2025-12-01 · stale |
| 26 | DeepSeek-Coder-V2-Lite-Instructopen ↗ | – | – | – | – | 128k | 2024-06-13 | 2024-06-17 · stale |
| 27 | Deepseek Coder 1 3B Base | – | – | – | – | – | 2023-11-02 | – |
| 28 | DeepSeek Coder 1.3B | – | – | – | – | – | – | – |
| 29 | DeepSeek Coder 33B | – | – | – | – | – | – | – |
| 30 | Deepseek Coder 33B Base | – | – | – | – | – | 2023-11-02 | – |
| 31 | Deepseek Coder 6 7B Base | – | – | – | – | – | 2023-11-02 | – |
| 32 | DeepSeek Coder 6.7B | – | – | – | – | – | – | – |
| 33 | Deepseek v3 Base | – | – | – | – | – | 2024-12-26 | – |
| 34 | DeepSeek-Coder-V2 236B | – | – | – | – | – | 2024-06-17 | – |
| 35 | DeepSeek-Coder-V2-Lite-Base | – | – | – | – | – | 2024-06-13 | – |
Latest news
- China's MiniMax completes export registration for open-source, overseas AI model licensingReuters · 23 Oct 2026
- Moonshot Debuts First AI Upgrade Since Its Own DeepSeek MomentBloomberg.com · 23 Oct 2026
- The Huawei-DeepSeek SymbiosisSemiAnalysis · 21 Oct 2026
- China’s DeepSeek Raises $15 Billion to Challenge Top US AI LabsBloomberg.com · 22 Oct 2026
- DeepSeek wrapping up $15 billion fundraising to fuel AI race, sources sayReuters · 22 Oct 2026
- DeepSeek reportedly raises $15B days after US restrictionsTechCrunch · 22 Oct 2026
- Deepseek's latest AI model is smarter than ever, but its low running costs come at a price — uses only one-quarter of the HBM and one-eighth of SSD storage but still battles hallucinations in agentic workloadsTom's Hardware · 23 Oct 2026
- ‘Jensen’s nightmare’: Huawei, DeepSeek narrow AI chip ecosystem gap with Nvidia: reportSouth China Morning Post · 22 Oct 2026