Model comparison
DeepSeek-V3.2-Exp vs Llama 3.1-70B
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.3 to 29.6 on the Noometry Index.
Last verified . 25 shared benchmarks.
DeepSeek
MetaSummary
- They share 25 benchmarks with published results for both. DeepSeek-V3.2-Exp scores higher in 9 categories and Llama 3.1-70B in 0 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-V3.2-Exp leads 41.7 to 13.5.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 87.8% for DeepSeek-V3.2-Exp and 3.6% for Llama 3.1-70B.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $0.40 / $0.40 for Llama 3.1-70B.
- DeepSeek-V3.2-Exp accepts more context: 164K tokens versus 128K.
Side by side
| DeepSeek-V3.2-Exp | Llama 3.1-70B | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 44.3 | 29.6 |
| Released | 2025-09-29 | 2024-07-23 |
| Weights | Open | Open |
| Context window | 164K | 128K |
| Max output | 66K | 4K |
| Input $ / M tokens | $0.26 | $0.40 |
| Output $ / M tokens | $0.38 | $0.40 |
| Results tracked | 49 | 35 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 46.5 (#65), Llama 3.1-70B: 30.3 (#296)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| WeirdML | 39.5% | 9% |
| LMArena Coding | 1454 | 1260 |
| SWE-bench Verified (bash only) | 70% | — |
| Aider Polyglot | 74.2% | — |
| LMArena WebDev | 1362 | — |
| SWE-bench Multilingual | 59% | — |
| SciCode | 38.9% | — |
| BigCodeBench Instruct | — | 46.1% |
| BigCodeBench Complete | — | 54.8% |
Agentic & Tool Use DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 32.7 (#59), Llama 3.1-70B: 25.1 (#112)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| TheAgentCompany | 42.9% | 6.9% |
| Terminal-Bench | 39.6% | — |
| APEX-Agents | 21.3% | — |
| Berkeley Function Calling Leaderboard | 56.7% | — |
| BALROG | — | 27.9% |
| Vending-Bench 2 | 1,034 | — |
Reasoning Too close to call
DeepSeek-V3.2-Exp: 22.1 (#208), Llama 3.1-70B: 21.6 (#220)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| LMArena Hard Prompts | 1434 | 1241 |
| DTBench | 87.7% | 60% |
| LMCA | 29.1% | 14.8% |
| Epoch Capabilities Index | 146.27 | 125.92 |
| ARC-AGI-2 | 4% | — |
| Kagi LLM Benchmark | 52.2% | — |
| NYT Connections (extended) | 36.7% | — |
| ARC-AGI-1 | 57% | — |
| CritPt | 2.9% | — |
| Chess Puzzles | 14% | — |
| Thematic Generalization | 65% | — |
Math DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 41.7 (#87), Llama 3.1-70B: 13.5 (#304)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 87.8% | 3.6% |
| LMArena Math | 1435 | 1252 |
| MathArena Final-Answer Competitions | 57.7% | — |
| ProofBench | 8% | — |
| Omni-MATH | — | 21% |
| MATH Level 5 | — | 36.7% |
| FrontierMath (Feb 2025 set) | 22.1% | — |
| FrontierMath Tier 4 (v1) | 2.1% | — |
Knowledge DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 51.7 (#66), Llama 3.1-70B: 24.2 (#269)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| GPQA Diamond | 83.4% | 44.2% |
| LMArena Expert | 1436 | 1209 |
| MMLU-Pro | — | 65.3% |
| Vectara Hallucination Rate | 5.3% | — |
| GPQA (HELM) | — | 42.6% |
| MMLU | — | 80.1% |
Multilingual DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 52.2 (#90), Llama 3.1-70B: 38.8 (#225)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| LMArena Non-English | 1409 | 1219 |
| LMArena Chinese | 1461 | 1215 |
| LMArena French | 1433 | 1261 |
| LMArena German | 1440 | 1222 |
| LMArena Japanese | 1374 | 1132 |
| LMArena Korean | 1371 | 1140 |
| LMArena Russian | 1424 | 1234 |
| LMArena Spanish | 1440 | 1253 |
Instruction Following DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 74.5 (#93), Llama 3.1-70B: 65.3 (#223)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| LMArena Instruction Following | 1413 | 1231 |
| IFEval | — | 82.1% |
Long Context DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 47.6 (#16), Llama 3.1-70B: 37.6 (#214)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| LMArena Longer Query | 1428 | 1241 |
| Fiction.LiveBench | 83.3% | — |
| CL-bench | 13.2% | — |
| CL-bench Life | 9.5% | — |
Writing & Preference DeepSeek-V3.2-Exp leads
DeepSeek-V3.2-Exp: 62.4 (#77), Llama 3.1-70B: 35.4 (#267)
| Benchmark | DeepSeek-V3.2-Exp | Llama 3.1-70B |
|---|---|---|
| LMArena Text | 1425 | 1261 |
| LMArena Creative Writing | 1403 | 1232 |
| EQ-Bench Creative Writing | 1515 | 784 |
| LMArena Multi-Turn | 1427 | 1256 |
| WildBench | — | 75.8% |
Frequently asked questions
Is DeepSeek-V3.2-Exp better than Llama 3.1-70B?
DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.3 to 29.6 on the Noometry Index.
Which is cheaper, DeepSeek-V3.2-Exp or Llama 3.1-70B?
DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; Llama 3.1-70B lists at $0.40 and $0.40.
Is DeepSeek-V3.2-Exp or Llama 3.1-70B better for coding?
DeepSeek-V3.2-Exp scores higher on coding benchmarks: 46.5 versus 30.3 in the Noometry coding category.
Which has the bigger context window?
DeepSeek-V3.2-Exp does, with 164K tokens against 128K.
How many benchmarks do DeepSeek-V3.2-Exp and Llama 3.1-70B share?
25 benchmarks have published results for both models. DeepSeek-V3.2-Exp has 49 scored results on Noometry and Llama 3.1-70B has 35.