Model comparison
DeepSeek V4 Pro vs Llama 3.2 3B
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 28.9 on the Noometry Index. Llama 3.2 3B costs 8.3× less per token, which makes it the better buy when DeepSeek V4 Pro's lead doesn't matter for your workload.
Last verified . 14 shared benchmarks.
DeepSeek
MetaSummary
- They share 14 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 9 categories and Llama 3.2 3B in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in writing & preference, where DeepSeek V4 Pro leads 65.5 to 24.7.
- Llama 3.2 3B is cheaper at $0.05 / $0.33 per million input/output tokens, against $0.66 / $1.98 for DeepSeek V4 Pro.
- DeepSeek V4 Pro accepts more context: 1M tokens versus 131K.
Side by side
| DeepSeek V4 Pro | Llama 3.2 3B | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 54.3 | 28.9 |
| Released | 2026-04-24 | 2024-09-24 |
| Weights | Open | Open |
| Context window | 1M | 131K |
| Max output | 393K | 118K |
| Input $ / M tokens | $0.66 | $0.05 |
| Output $ / M tokens | $1.98 | $0.33 |
| Results tracked | 48 | 18 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek V4 Pro leads
DeepSeek V4 Pro: 52.4 (#34), Llama 3.2 3B: 27.6 (#319)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Coding | 1470 | 1098 |
| SWE-bench Verified | 77.6% | — |
| FrontierCode | 28.6% | — |
| LMArena WebDev | 1582 | — |
| SciCode | 51% | — |
| WeirdML | 66.2% | — |
| BigCodeBench Instruct | — | 23.4% |
| BigCodeBench Complete | — | 28.3% |
| ALE-Bench | 1,403 | — |
Agentic & Tool Use DeepSeek V4 Pro leads
DeepSeek V4 Pro: 32.8 (#58), Llama 3.2 3B: 20.1 (#143)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| APEX-Agents | 47.3% | — |
| Berkeley Function Calling Leaderboard | — | 21.9% |
| BALROG | — | 10.1% |
| Vending-Bench 2 | 3,285 | — |
Reasoning DeepSeek V4 Pro leads
DeepSeek V4 Pro: 56.5 (#24), Llama 3.2 3B: 21.0 (#228)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Hard Prompts | 1461 | 1095 |
| ARC-AGI-2 | 61.3% | — |
| Kagi LLM Benchmark | 53.5% | — |
| NYT Connections (extended) | 91.3% | — |
| ARC-AGI-1 | 90.5% | — |
| CritPt | 18% | — |
| Chess Puzzles | 47% | — |
| Mystery Game Puzzles | 43% | — |
| DTBench | 93.9% | — |
| LMCA | 45.5% | — |
| Surface Evolver Bench | 40% | — |
| Epoch Capabilities Index | 155.31 | — |
| ForecastBench | 56.1 | — |
Math DeepSeek V4 Pro leads
DeepSeek V4 Pro: 64.8 (#30), Llama 3.2 3B: 32.4 (#214)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Math | 1455 | 1126 |
| FrontierMath (Tiers 1-3) | 64.6% | — |
| FrontierMath Tier 4 | 26.8% | — |
| MathArena Final-Answer Competitions | 76.6% | — |
| OTIS Mock AIME 2024-2025 | 98.6% | — |
| ProofBench | 50% | — |
Knowledge DeepSeek V4 Pro leads
DeepSeek V4 Pro: 59.5 (#31), Llama 3.2 3B: 29.7 (#235)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Expert | 1464 | 1090 |
| GPQA Diamond | 91.7% | — |
| SimpleQA Verified | 52.9% | — |
| Vectara Hallucination Rate | 8.6% | — |
Multilingual DeepSeek V4 Pro leads
DeepSeek V4 Pro: 54.4 (#45), Llama 3.2 3B: 26.2 (#281)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Non-English | 1439 | 1019 |
| LMArena Chinese | 1486 | 1017 |
| LMArena German | 1458 | 1056 |
| LMArena Russian | 1453 | 949 |
| LMArena French | 1472 | — |
| LMArena Japanese | 1445 | — |
| LMArena Korean | 1447 | — |
| LMArena Spanish | 1458 | — |
Instruction Following DeepSeek V4 Pro leads
DeepSeek V4 Pro: 76.1 (#47), Llama 3.2 3B: 56.0 (#275)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Instruction Following | 1448 | 1089 |
Long Context DeepSeek V4 Pro leads
DeepSeek V4 Pro: 45.0 (#51), Llama 3.2 3B: 33.4 (#261)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Longer Query | 1458 | 1100 |
| CL-bench Life | 13.5% | — |
Writing & Preference DeepSeek V4 Pro leads
DeepSeek V4 Pro: 65.5 (#46), Llama 3.2 3B: 24.7 (#307)
| Benchmark | DeepSeek V4 Pro | Llama 3.2 3B |
|---|---|---|
| LMArena Text | 1451 | 1110 |
| LMArena Creative Writing | 1446 | 1094 |
| EQ-Bench Creative Writing | 1553 | 595 |
| LMArena Multi-Turn | 1467 | 1105 |
| EQ-Bench 4 | 1166 | — |
Frequently asked questions
Is DeepSeek V4 Pro better than Llama 3.2 3B?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 28.9 on the Noometry Index. Llama 3.2 3B costs 8.3× less per token, which makes it the better buy when DeepSeek V4 Pro's lead doesn't matter for your workload.
Which is cheaper, DeepSeek V4 Pro or Llama 3.2 3B?
Llama 3.2 3B is cheaper. It lists at $0.05 per million input tokens and $0.33 per million output tokens; DeepSeek V4 Pro lists at $0.66 and $1.98.
Is DeepSeek V4 Pro or Llama 3.2 3B better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 27.6 in the Noometry coding category.
Which has the bigger context window?
DeepSeek V4 Pro does, with 1M tokens against 131K.
How many benchmarks do DeepSeek V4 Pro and Llama 3.2 3B share?
14 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and Llama 3.2 3B has 18.