Model comparison
DeepSeek-V3.1-Terminus vs DeepSeek-V3.2-Exp
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 42.6 on the Noometry Index.
Last verified . 15 shared benchmarks.
DeepSeekSummary
- They share 15 benchmarks with published results for both. DeepSeek-V3.1-Terminus scores higher in 1 category and DeepSeek-V3.2-Exp in 6 categories; 5 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-V3.2-Exp leads 40.8 to 35.7.
- The biggest single-benchmark swing is DTBench: 81.3% for DeepSeek-V3.1-Terminus and 87.7% for DeepSeek-V3.2-Exp.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $0.30 / $1 for DeepSeek-V3.1-Terminus.
Side by side
| DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp | |
|---|---|---|
| Provider | DeepSeek | DeepSeek |
| Noometry Index | 42.6 | 44.9 |
| Released | 2025-09-22 | 2025-09-29 |
| Weights | Open | Open |
| Context window | 164K | 164K |
| Max output | 66K | 66K |
| Input $ / M tokens | $0.30 | $0.26 |
| Output $ / M tokens | $1 | $0.38 |
| Results tracked | 16 | 48 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.2-Exp leads
DeepSeek-V3.1-Terminus: 41.5 (#114), DeepSeek-V3.2-Exp: 46.4 (#61)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| SciCode | 40.6% | 38.9% |
| LMArena Coding | 1426 | 1454 |
| SWE-bench Verified (bash only) | — | 70% |
| Aider Polyglot | — | 74.2% |
| LMArena WebDev | — | 1362 |
| SWE-bench Multilingual | — | 59% |
| WeirdML | — | 39.5% |
| ALE-Bench | 745.17 | — |
Agentic & Tool Use Not comparable
DeepSeek-V3.1-Terminus: —, DeepSeek-V3.2-Exp: 37.9 (#31)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| Terminal-Bench | — | 39.6% |
| Berkeley Function Calling Leaderboard | — | 56.7% |
| TheAgentCompany | — | 42.9% |
| Vending-Bench 2 | — | 1,034 |
Reasoning DeepSeek-V3.1-Terminus leads
DeepSeek-V3.1-Terminus: 25.7 (#126), DeepSeek-V3.2-Exp: 22.0 (#185)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| Kagi LLM Benchmark | 57.4% | 52.2% |
| CritPt | 1.7% | 2.9% |
| LMArena Hard Prompts | 1427 | 1433 |
| DTBench | 81.3% | 87.7% |
| LMCA | 28.6% | 29.1% |
| ARC-AGI-2 | — | 4% |
| NYT Connections (extended) | — | 36.7% |
| ARC-AGI-1 | — | 57% |
| Chess Puzzles | — | 14% |
| Thematic Generalization | — | 65% |
| Epoch Capabilities Index | — | 146.28 |
Math DeepSeek-V3.2-Exp leads
DeepSeek-V3.1-Terminus: 35.7 (#138), DeepSeek-V3.2-Exp: 40.8 (#75)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Math | 1402 | 1435 |
| MathArena Final-Answer Competitions | — | 57.7% |
| OTIS Mock AIME 2024-2025 | — | 87.8% |
| ProofBench | — | 8% |
| FrontierMath (Feb 2025 set) | — | 22.1% |
| FrontierMath Tier 4 (v1) | — | 2.1% |
Knowledge Not comparable
DeepSeek-V3.1-Terminus: —, DeepSeek-V3.2-Exp: 51.6 (#61)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| GPQA Diamond | — | 83.4% |
| Vectara Hallucination Rate | — | 5.3% |
| LMArena Expert | — | 1436 |
Multilingual Too close to call
DeepSeek-V3.1-Terminus: 52.3 (#87), DeepSeek-V3.2-Exp: 52.4 (#85)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Non-English | 1407 | 1409 |
| LMArena Russian | 1434 | 1423 |
| LMArena Chinese | — | 1461 |
| LMArena French | — | 1435 |
| LMArena German | — | 1440 |
| LMArena Japanese | — | 1377 |
| LMArena Korean | — | 1372 |
| LMArena Spanish | — | 1441 |
Instruction Following Too close to call
DeepSeek-V3.1-Terminus: 74.4 (#96), DeepSeek-V3.2-Exp: 74.7 (#89)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Instruction Following | 1406 | 1413 |
Long Context DeepSeek-V3.2-Exp leads
DeepSeek-V3.1-Terminus: 43.5 (#88), DeepSeek-V3.2-Exp: 47.6 (#15)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Longer Query | 1426 | 1428 |
| Fiction.LiveBench | — | 83.3% |
| CL-bench | — | 13.2% |
| CL-bench Life | — | 9.5% |
Writing & Preference DeepSeek-V3.2-Exp leads
DeepSeek-V3.1-Terminus: 61.3 (#87), DeepSeek-V3.2-Exp: 62.5 (#73)
| Benchmark | DeepSeek-V3.1-Terminus | DeepSeek-V3.2-Exp |
|---|---|---|
| LMArena Text | 1420 | 1425 |
| LMArena Creative Writing | 1404 | 1403 |
| LMArena Multi-Turn | 1411 | 1427 |
| EQ-Bench Creative Writing | — | 1515 |
Frequently asked questions
Is DeepSeek-V3.1-Terminus better than DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 42.6 on the Noometry Index.
Which is cheaper, DeepSeek-V3.1-Terminus or DeepSeek-V3.2-Exp?
DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; DeepSeek-V3.1-Terminus lists at $0.30 and $1.
Is DeepSeek-V3.1-Terminus or DeepSeek-V3.2-Exp better for coding?
DeepSeek-V3.2-Exp scores higher on coding benchmarks: 46.4 versus 41.5 in the Noometry coding category.
Which has the bigger context window?
Both accept 164K tokens.
How many benchmarks do DeepSeek-V3.1-Terminus and DeepSeek-V3.2-Exp share?
15 benchmarks have published results for both models. DeepSeek-V3.1-Terminus has 16 scored results on Noometry and DeepSeek-V3.2-Exp has 48.