Reasoning
Mathematical and scientific problem solving.
Model profile / DeepSeek
DeepSeek V3.2 is a open-weight model tracked via Artificial Analysis independent evaluations for capability comparison across coding, agents, and reasoning.
Four areas of evidence
Mathematical and scientific problem solving.
Software engineering and programming.
Factual reliability and instruction following.
Tool use, workflows and professional tasks.
Current index snapshot · 2026-09-16. Each category uses its own scale; scores across categories are not directly comparable. Evidence counts describe the retained Capabilities inputs; a family can contain several tasks or configurations. A point lead does not establish superiority.
Performance in context
DeepSeek V3.2 is highlighted wherever a compatible measurement is available.
DeepSeek V3.2: Retired or changed served checkpoint; current DeepSeek endpoint prices cannot transfer to frozen predecessor weights
Lumina Capabilities Index
The current Capabilities cohort, with this model highlighted.
Chart loads as you explore
DeepSeek V3.2: No exact base-profile release and documented 10K output measurement; other effort/protocol/alias records not substituted
Lumina Capabilities Index
The current Capabilities cohort, with this model highlighted.
Chart loads as you explore
Explore the detail
Score uses the admitted direct native evidence under the selected method.
| Benchmark | Result | Published configuration | Source |
|---|---|---|---|
| GPQA Diamond2026 | 75.1% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 21 July 2026 ↗Observed 2026-07-21 · checked 2026-07-21source-checked · reference-only |
| GPQA Diamond2026 | 75.1% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 2026-07-27 ↗Observed 2026-07-27 · checked 2026-07-27source-checked · reference-only |
| GPQA Diamond2026 | 75.1% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 2026-08-01 ↗Observed 2026-08-01 · checked 2026-08-01source-checked · reference-only |
| Humanity's Last Exam2026 | 10.5% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 21 July 2026 ↗Observed 2026-07-21 · checked 2026-07-21source-checked · reference-only |
| Humanity's Last Exam2026 | 10.5% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 2026-07-27 ↗Observed 2026-07-27 · checked 2026-07-27source-checked · reference-only |
| Humanity's Last Exam2026 | 10.5% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 2026-08-01 ↗Observed 2026-08-01 · checked 2026-08-01source-checked · reference-only |
| IFBench2026 | 49% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 21 July 2026 ↗Observed 2026-07-21 · checked 2026-07-21source-checked · reference-only |
| IFBench2026 | 49% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 2026-07-27 ↗Observed 2026-07-27 · checked 2026-07-27source-checked · reference-only |
| IFBench2026 | 49% | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 2026-08-01 ↗Observed 2026-08-01 · checked 2026-08-01source-checked · reference-only |
| AA-Omniscience2026 | -46.7 index | Exact BenchLM registry variant DeepSeek V3.2; bulk export does not retain a complete upstream harness configuration.Source-native system | BenchLM public datasets — 21 July 2026 ↗Observed 2026-07-21 · checked 2026-07-21source-checked · reference-only |