Model comparison
DeepSeek-V3.2-Exp vs Hy4 preview
Hy4 preview is the stronger model overall, scoring 45.3 to 44.3 on the Noometry Index. DeepSeek-V3.2-Exp costs 3.9× less per token, which makes it the better buy when Hy4 preview's lead doesn't matter for your workload.
Last verified . 3 shared benchmarks.
DeepSeek
TencentSummary
- They share 3 benchmarks with published results for both. DeepSeek-V3.2-Exp scores higher in 0 categories and Hy4 preview in 3 categories; 3 gaps are clear of the uncertainty.
- The widest gap is in math, where Hy4 preview leads 55.7 to 41.7.
- The biggest single-benchmark swing is ProofBench: 8% for DeepSeek-V3.2-Exp and 75% for Hy4 preview.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $0.75 / $2.25 for Hy4 preview.
- Hy4 preview accepts more context: 1.05M tokens versus 164K.
Side by side
| DeepSeek-V3.2-Exp | Hy4 preview | |
|---|---|---|
| Provider | DeepSeek | Tencent |
| Noometry Index | 44.3 | 45.3 |
| Released | 2025-09-29 | 2026-08-28 |
| Weights | Open | Open |
| Context window | 164K | 1.05M |
| Max output | 66K | 64K |
| Input $ / M tokens | $0.26 | $0.75 |
| Output $ / M tokens | $0.38 | $2.25 |
| Results tracked | 49 | 3 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding Hy4 preview leads
DeepSeek-V3.2-Exp: 46.5 (#65), Hy4 preview: 51.6 (#38)
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| LMArena WebDev | 1362 | 1632 |
| SWE-bench Verified (bash only) | 70% | — |
| Aider Polyglot | 74.2% | — |
| SWE-bench Multilingual | 59% | — |
| SciCode | 38.9% | — |
| WeirdML | 39.5% | — |
| LMArena Coding | 1454 | — |
Agentic & Tool Use Not comparable
DeepSeek-V3.2-Exp: 32.7 (#59), Hy4 preview: —
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| Terminal-Bench | 39.6% | — |
| APEX-Agents | 21.3% | — |
| Berkeley Function Calling Leaderboard | 56.7% | — |
| TheAgentCompany | 42.9% | — |
| Vending-Bench 2 | 1,034 | — |
Reasoning Hy4 preview leads
DeepSeek-V3.2-Exp: 22.1 (#208), Hy4 preview: 31.9 (#79)
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| NYT Connections (extended) | 36.7% | 68.2% |
| ARC-AGI-2 | 4% | — |
| Kagi LLM Benchmark | 52.2% | — |
| ARC-AGI-1 | 57% | — |
| CritPt | 2.9% | — |
| Chess Puzzles | 14% | — |
| Thematic Generalization | 65% | — |
| LMArena Hard Prompts | 1434 | — |
| DTBench | 87.7% | — |
| LMCA | 29.1% | — |
| Epoch Capabilities Index | 146.27 | — |
Math Hy4 preview leads
DeepSeek-V3.2-Exp: 41.7 (#87), Hy4 preview: 55.7 (#42)
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| ProofBench | 8% | 75% |
| MathArena Final-Answer Competitions | 57.7% | — |
| OTIS Mock AIME 2024-2025 | 87.8% | — |
| LMArena Math | 1435 | — |
| FrontierMath (Feb 2025 set) | 22.1% | — |
| FrontierMath Tier 4 (v1) | 2.1% | — |
Knowledge Not comparable
DeepSeek-V3.2-Exp: 51.7 (#66), Hy4 preview: —
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| GPQA Diamond | 83.4% | — |
| Vectara Hallucination Rate | 5.3% | — |
| LMArena Expert | 1436 | — |
Multilingual Not comparable
DeepSeek-V3.2-Exp: 52.2 (#90), Hy4 preview: —
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| LMArena Non-English | 1409 | — |
| LMArena Chinese | 1461 | — |
| LMArena French | 1433 | — |
| LMArena German | 1440 | — |
| LMArena Japanese | 1374 | — |
| LMArena Korean | 1371 | — |
| LMArena Russian | 1424 | — |
| LMArena Spanish | 1440 | — |
Instruction Following Not comparable
DeepSeek-V3.2-Exp: 74.5 (#93), Hy4 preview: —
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| LMArena Instruction Following | 1413 | — |
Long Context Not comparable
DeepSeek-V3.2-Exp: 47.6 (#16), Hy4 preview: —
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| Fiction.LiveBench | 83.3% | — |
| CL-bench | 13.2% | — |
| CL-bench Life | 9.5% | — |
| LMArena Longer Query | 1428 | — |
Writing & Preference Not comparable
DeepSeek-V3.2-Exp: 62.4 (#77), Hy4 preview: —
| Benchmark | DeepSeek-V3.2-Exp | Hy4 preview |
|---|---|---|
| LMArena Text | 1425 | — |
| LMArena Creative Writing | 1403 | — |
| EQ-Bench Creative Writing | 1515 | — |
| LMArena Multi-Turn | 1427 | — |
Frequently asked questions
Is DeepSeek-V3.2-Exp better than Hy4 preview?
Hy4 preview is the stronger model overall, scoring 45.3 to 44.3 on the Noometry Index. DeepSeek-V3.2-Exp costs 3.9× less per token, which makes it the better buy when Hy4 preview's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3.2-Exp or Hy4 preview?
DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; Hy4 preview lists at $0.75 and $2.25.
Is DeepSeek-V3.2-Exp or Hy4 preview better for coding?
Hy4 preview scores higher on coding benchmarks: 51.6 versus 46.5 in the Noometry coding category.
Which has the bigger context window?
Hy4 preview does, with 1.05M tokens against 164K.
How many benchmarks do DeepSeek-V3.2-Exp and Hy4 preview share?
3 benchmarks have published results for both models. DeepSeek-V3.2-Exp has 49 scored results on Noometry and Hy4 preview has 3.