Tier
Major
Headquarters
Hangzhou, China
Founded
2023
Flagship
DeepSeek R1 / V3
Models tracked
31
Artificial Analysis
Top intelligence
44.3
Artificial Analysis Index
Cheapest model
$0.00
blended /1M
Fastest output
118
tokens / sec
Flagship Model
DeepSeek’s top-scoring model on the Artificial Analysis Intelligence Index.
Most intelligent model
DeepSeek V4 Pro (Reasoning, Max Effort)
Released 2026-04-24
Intelligence
44.3
Coding
59.4
Math
—
In $/1M
$0.43
Out $/1M
$0.87
Speed t/s
71
Benchmark breakdown
Models & benchmarks
Every DeepSeek model on file — intelligence, coding, price, and speed.
| Model | Released | Intel | Coding | Math | In $/1M | Out $/1M | Speed t/s | Latency s |
|---|---|---|---|---|---|---|---|---|
| DeepSeek V4 Pro (Reasoning, Max Effort) | 2026-04-24 | 44.3 | 59.4 | — | $0.43 | $0.87 | 71 | 0.91 |
| DeepSeek V4 Pro (Reasoning, High Effort) | 2026-04-24 | 43.1 | 58.7 | — | $0.43 | $0.87 | 73 | 0.96 |
| DeepSeek V4 Flash (Reasoning, Max Effort) | 2026-04-24 | 40.3 | 56.2 | — | $0.14 | $0.28 | 118 | 0.90 |
| DeepSeek V4 Flash (Reasoning, High Effort) | 2026-04-24 | 37.5 | 52.0 | — | $0.14 | $0.28 | 0 | 0.00 |
| DeepSeek V3.2 (Reasoning) | 2025-12-01 | 32.0 | 44.2 | 92.0 | $0.28 | $0.42 | 0 | 0.00 |
| DeepSeek V4 Pro (Non-reasoning) | 2026-04-24 | 31.2 | — | — | $0.43 | $0.87 | 69 | 1.04 |
| DeepSeek V3.1 Terminus (Reasoning) | 2025-09-22 | 30.4 | 43.5 | 89.7 | $1.64 | $2.75 | 0 | 0.00 |
| DeepSeek V4 Flash (Non-reasoning) | 2026-04-24 | 28.7 | — | — | $0.14 | $0.28 | 116 | 0.91 |
| DeepSeek V3.2 Exp (Reasoning) | 2025-09-29 | 25.4 | — | 87.7 | $0.28 | $0.42 | 0 | 0.00 |
| DeepSeek V3.2 (Non-reasoning) | 2025-12-01 | 24.7 | — | 59.0 | $0.28 | $0.42 | 0 | 0.00 |
| DeepSeek V3.2 Speciale | 2025-12-01 | 22.2 | — | 96.7 | $0.00 | $0.00 | 0 | 0.00 |
| DeepSeek V3.1 Terminus (Non-reasoning) | 2025-09-22 | 21.4 | — | 53.7 | $0.27 | $1.00 | 0 | 0.00 |
| DeepSeek V3.2 Exp (Non-reasoning) | 2025-09-29 | 21.3 | — | 57.7 | $0.28 | $0.42 | 0 | 0.00 |
| DeepSeek V3.1 (Non-reasoning) | 2025-08-21 | 21.0 | — | 49.7 | $0.56 | $1.68 | 0 | 0.00 |
Trajectory
DeepSeek’s model releases over time, by intelligence.
House strengths
Where DeepSeek’s lineup over-indexes, averaged across its models.
Capability profile
Best value
DeepSeek V4 Flash (Reasoning, Max Effort)
Most intelligence per dollar — 40.3 index at $0.17/1M blended.
230 index points per $1
Governance
DeepSeek’s license posture and openness across its lineup.
Open weight
23
of 23 models
Proprietary
0
of 23 models
Commercial-OK
17
license permits
Modalities
Text · Image
Via LLM Stats
Media models
Image, video, and speech models, ranked by arena Elo.
| Model | Type | Released | Elo | Rank |
|---|---|---|---|---|
| Janus Pro | text to image | 2025-01 | 712 | #145 |