DeepSeek V3.2
| Released | 2025-12-01 |
|---|---|
| Price per 1M | $0.29 in · $0.43 out |
| Overall score | 50.6 · rank 106 |
| Context window | 128,000 tokens |
| Provider | DeepSeek |
| Weights | Open weights (MIT License) |
Scores
| Dimension | Score | Rank | Benchmarks |
|---|---|---|---|
| Overall: Balanced | 50.6 | 106 | 20 |
| Overall: Coding | 49.7 | 112 | 10 |
| Overall: Reasoning | 50.9 | 105 | 9 |
| Overall: Agentic work | 51.0 | 108 | 8 |
| Overall: Frontend | 42.9 | 159 | 4 |
| Coding | 52.5 | 99 | 2 |
| Agents | 50.4 | 102 | 2 |
| Reasoning | 50.5 | 110 | 4 |
| Math | 51.5 | 99 | 4 |
| Knowledge | 50.8 | 87 | 1 |
| Frontend | 38.2 | 114 | 2 |
| General | 51.6 | 90 | 5 |
| Value vs price | +6.2 | 62 |
Configurations
| Configuration | Balanced | Benchmarks |
|---|---|---|
| Unspecified setting | 49.1 ±1.6 | 15 |
| High | 50.1 ±4.5 | 3 |
| Non-reasoning | 47.6 ±5.2 | 3 |
| Thinking | 51.0 ±3.6 | 5 |
Evidence
Coding 52.5 ±3.3
| Benchmark | Result | Source | Date |
|---|---|---|---|
| SWE-bench Verified (Vals run) | 67.6% | Vals AI | 2026-09-01 |
| LiveCodeBench (Vals run) | 80.7% | Vals AI | 2026-09-01 |
Agents 50.4 ±1.6
| Benchmark | Result | Source | Date |
|---|---|---|---|
| APEX-Agents | 21.3% | Epoch AI | 2026-11-11 |
| Terminal-Bench 2.0 | 39.6% | Epoch AI | 2025-12-01 |
Reasoning 50.5 ±1.9
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Extended NYT Connections | 36.7% | Lech Mazur | 2026-11-10 |
| ARC-AGI-2 | 4.0% | Epoch AI | 2026-11-11 |
| ARC-AGI-2 (ARC Prize verified) | 4.0% | ARC Prize | 2026-11-11 |
| GPQA Diamond (Vals run) | 80.3% | Vals AI | 2026-09-01 |
| ARC-AGI-1 (ARC Prize verified) | 57.0% | ARC Prize | 2026-11-11 |
| ARC-AGI-1 | 57.0% | Epoch AI | 2026-11-11 |
Math 51.5 ±1.8
| Benchmark | Result | Source | Date |
|---|---|---|---|
| FrontierMath Tiers 1-3 (2025 set) | 22.1% | Epoch AI | 2025-12-22 |
| ProofBench | 8.0% | Epoch AI | 2026-11-11 |
| FrontierMath Tier 4 (2025 set) | 2.1% | Epoch AI | 2025-12-16 |
| MathArena Final-Answer Competitions | 57.7% | MathArena | 2026-11-11 |
Knowledge 50.8 ±4.7
| Benchmark | Result | Source | Date |
|---|---|---|---|
| MMLU-Pro (Vals run) | 84.9% | Vals AI | 2026-09-01 |
Frontend 38.2 ±0.7
| Benchmark | Result | Source | Date |
|---|---|---|---|
| Vibe Code Bench v1.1 | 5.1% | Vals AI | 2026-11-09 |
| LMArena WebDev | 1,363 | LMArena | 2026-11-09 |
General 51.6 ±1.8
| Benchmark | Result | Source | Date |
|---|---|---|---|
| LLM Debate Benchmark | 1,355 | Lech Mazur | 2026-10-26 |
| EQ-Bench Creative Writing v3 | 1,515 | EQ-Bench | 2026-11-11 |
| LMArena Text | 1,425 | LMArena | 2026-11-09 |
| Creative Story-Writing (Lech Mazur) | 13.0% | Lech Mazur | 2026-10-23 |
| EQ-Bench Longform Creative Writing | 66.8% | EQ-Bench | 2026-11-11 |
Results not used in the score
| Benchmark | Result | Source | Why not used |
|---|---|---|---|
| Extended NYT Connections | 8.2% | Lech Mazur | A stronger configuration result was used |
| GPQA Diamond (Vals run) | 76.3% | Vals AI | A stronger configuration result was used |
| LMArena Text | 1,423 | LMArena | A stronger configuration result was used |
| LMArena WebDev | 1,326 | LMArena | A stronger configuration result was used |
| LiveCodeBench (Vals run) | 69.9% | Vals AI | A stronger configuration result was used |
| MMLU-Pro (Vals run) | 83.1% | Vals AI | A stronger configuration result was used |
Names matched to this model
deepseek-v3-2, deepseek-v3p2