Model comparison
DeepSeek V4 Pro vs Llama 3.1-8B
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 23.0 on the Noometry Index. Llama 3.1-8B costs 17× less per token, which makes it the better buy when DeepSeek V4 Pro's lead doesn't matter for your workload.
Last verified . 27 shared benchmarks.
DeepSeek
MetaSummary
- They share 27 benchmarks with published results for both. DeepSeek V4 Pro scores higher in 9 categories and Llama 3.1-8B in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek V4 Pro leads 64.8 to 10.2.
- The biggest single-benchmark swing is OTIS Mock AIME 2024-2025: 98.6% for DeepSeek V4 Pro and 1.7% for Llama 3.1-8B.
- Llama 3.1-8B is cheaper at $0.05 / $0.08 per million input/output tokens, against $0.66 / $1.98 for DeepSeek V4 Pro.
- DeepSeek V4 Pro accepts more context: 1M tokens versus 128K.
Side by side
| DeepSeek V4 Pro | Llama 3.1-8B | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 54.3 | 23.0 |
| Released | 2026-04-24 | 2024-07-23 |
| Weights | Open | Open |
| Context window | 1M | 128K |
| Max output | 393K | 4K |
| Input $ / M tokens | $0.66 | $0.05 |
| Output $ / M tokens | $1.98 | $0.08 |
| Results tracked | 48 | 43 |
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 3.1-8B: 20.2 (#340)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| SciCode | 51% | 13.2% |
| WeirdML | 66.2% | 1.7% |
| LMArena Coding | 1470 | 1195 |
| SWE-bench Verified | 77.6% | — |
| FrontierCode | 28.6% | — |
| LMArena WebDev | 1582 | — |
| BigCodeBench Instruct | — | 32.8% |
| BigCodeBench Complete | — | 40.5% |
| ALE-Bench | 1,403 | — |
| HumanEval+ | — | 62.8% |
| MBPP+ | — | 55.6% |
Agentic & Tool Use DeepSeek V4 Pro leads
DeepSeek V4 Pro: 32.8 (#58), Llama 3.1-8B: 22.5 (#131)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| APEX-Agents | 47.3% | — |
| Berkeley Function Calling Leaderboard | — | 25.8% |
| BALROG | — | 15.1% |
| Vending-Bench 2 | 3,285 | — |
Reasoning DeepSeek V4 Pro leads
DeepSeek V4 Pro: 56.5 (#24), Llama 3.1-8B: 14.9 (#321)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| CritPt | 18% | 0% |
| Chess Puzzles | 47% | 0% |
| LMArena Hard Prompts | 1461 | 1175 |
| DTBench | 93.9% | 50.9% |
| LMCA | 45.5% | 5.4% |
| Epoch Capabilities Index | 155.31 | 116.57 |
| ARC-AGI-2 | 61.3% | — |
| Kagi LLM Benchmark | 53.5% | — |
| NYT Connections (extended) | 91.3% | — |
| ARC-AGI-1 | 90.5% | — |
| Mystery Game Puzzles | 43% | — |
| Surface Evolver Bench | 40% | — |
| ForecastBench | 56.1 | — |
| PIQA | — | 81.2% |
Math DeepSeek V4 Pro leads
DeepSeek V4 Pro: 64.8 (#30), Llama 3.1-8B: 10.2 (#317)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 98.6% | 1.7% |
| LMArena Math | 1455 | 1179 |
| FrontierMath (Tiers 1-3) | 64.6% | — |
| FrontierMath Tier 4 | 26.8% | — |
| MathArena Final-Answer Competitions | 76.6% | — |
| ProofBench | 50% | — |
| Omni-MATH | — | 13.7% |
| MATH Level 5 | — | 22.9% |
| GSM8K | — | 82.4% |
Knowledge DeepSeek V4 Pro leads
DeepSeek V4 Pro: 59.5 (#31), Llama 3.1-8B: 8.0 (#307)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| GPQA Diamond | 91.7% | 27% |
| LMArena Expert | 1464 | 1144 |
| SimpleQA Verified | 52.9% | — |
| MMLU-Pro | — | 40.6% |
| Vectara Hallucination Rate | 8.6% | — |
| GPQA (HELM) | — | 24.7% |
| BoolQ | — | 82.8% |
| MMLU | — | 56.1% |
Multilingual DeepSeek V4 Pro leads
DeepSeek V4 Pro: 54.4 (#45), Llama 3.1-8B: 34.0 (#249)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| LMArena Non-English | 1439 | 1148 |
| LMArena Chinese | 1486 | 1151 |
| LMArena French | 1472 | 1177 |
| LMArena German | 1458 | 1144 |
| LMArena Japanese | 1445 | 1061 |
| LMArena Korean | 1447 | 1053 |
| LMArena Russian | 1453 | 1158 |
| LMArena Spanish | 1458 | 1169 |
Instruction Following DeepSeek V4 Pro leads
DeepSeek V4 Pro: 76.1 (#47), Llama 3.1-8B: 58.9 (#258)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| LMArena Instruction Following | 1448 | 1159 |
| IFEval | — | 74.3% |
Long Context DeepSeek V4 Pro leads
DeepSeek V4 Pro: 45.0 (#51), Llama 3.1-8B: 35.8 (#238)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| LMArena Longer Query | 1458 | 1182 |
| CL-bench Life | 13.5% | — |
Writing & Preference DeepSeek V4 Pro leads
DeepSeek V4 Pro: 65.5 (#46), Llama 3.1-8B: 29.7 (#290)
| Benchmark | DeepSeek V4 Pro | Llama 3.1-8B |
|---|---|---|
| LMArena Text | 1451 | 1187 |
| LMArena Creative Writing | 1446 | 1154 |
| EQ-Bench Creative Writing | 1553 | 713 |
| LMArena Multi-Turn | 1467 | 1172 |
| WildBench | — | 68.7% |
| EQ-Bench 4 | 1166 | — |
Frequently asked questions
Is DeepSeek V4 Pro better than Llama 3.1-8B?
DeepSeek V4 Pro is the stronger model overall, scoring 54.3 to 23.0 on the Noometry Index. Llama 3.1-8B costs 17× 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 3.1-8B?
Llama 3.1-8B is cheaper. It lists at $0.05 per million input tokens and $0.08 per million output tokens; DeepSeek V4 Pro lists at $0.66 and $1.98.
Is DeepSeek V4 Pro or Llama 3.1-8B better for coding?
DeepSeek V4 Pro scores higher on coding benchmarks: 52.4 versus 20.2 in the Noometry coding category.
Which has the bigger context window?
DeepSeek V4 Pro does, with 1M tokens against 128K.
How many benchmarks do DeepSeek V4 Pro and Llama 3.1-8B share?
27 benchmarks have published results for both models. DeepSeek V4 Pro has 48 scored results on Noometry and Llama 3.1-8B has 43.