Model comparison
DeepSeek-V2.5 (Sep 2024) vs Llama 4 Maverick
DeepSeek-V2.5 (Sep 2024) is the stronger model overall, scoring 37.6 to 30.9 on the Noometry Index.
Last verified . 20 shared benchmarks.
DeepSeek
MetaSummary
- They share 20 benchmarks with published results for both. DeepSeek-V2.5 (Sep 2024) scores higher in 7 categories and Llama 4 Maverick in 1 category; 7 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek-V2.5 (Sep 2024) leads 25.6 to 10.1.
- The biggest single-benchmark swing is BigCodeBench Complete: 53.2% for DeepSeek-V2.5 (Sep 2024) and 61.4% for Llama 4 Maverick.
Side by side
| DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 37.6 | 30.9 |
| Released | 2024-09-06 | 2025-04-05 |
| Weights | Open | Open |
| Context window | — | 128K |
| Max output | — | 4K |
| Input $ / M tokens | — | $0.19 |
| Output $ / M tokens | — | $0.65 |
| Results tracked | 22 | 54 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 31.7 (#281), Llama 4 Maverick: 26.6 (#324)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| Aider Polyglot | 17.8% | 15.6% |
| BigCodeBench Instruct | 48.6% | 49.7% |
| LMArena Coding | 1309 | 1302 |
| BigCodeBench Complete | 53.2% | 61.4% |
| SWE-bench Verified (bash only) | — | 21% |
| SciCode | — | 33.1% |
| WeirdML | — | 24.5% |
| ALE-Bench | — | 172.97 |
| HumanEval+ | 83.5% | — |
| MBPP+ | 74.1% | — |
Agentic & Tool Use Not comparable
DeepSeek-V2.5 (Sep 2024): —, Llama 4 Maverick: 28.2 (#91)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 37.3% |
Reasoning DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 25.6 (#145), Llama 4 Maverick: 10.1 (#342)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Hard Prompts | 1289 | 1281 |
| ARC-AGI-2 | — | 0% |
| SimpleBench | — | 27.7% |
| Kagi LLM Benchmark | — | 55.9% |
| NYT Connections (extended) | — | 8% |
| ARC-AGI-1 | — | 4.4% |
| CritPt | — | 0% |
| EnigmaEval | — | 0.6% |
| DTBench | — | 61.9% |
| LMCA | — | 15.9% |
| Epoch Capabilities Index | — | 132.2 |
| ForecastBench | — | 57.5 |
Math DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 35.9 (#177), Llama 4 Maverick: 26.0 (#262)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Math | 1288 | 1299 |
| OTIS Mock AIME 2024-2025 | — | 20.6% |
| Omni-MATH | — | 42.2% |
| MATH Level 5 | — | 73% |
| FrontierMath (Feb 2025 set) | — | 0.7% |
Knowledge DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 34.8 (#193), Llama 4 Maverick: 33.4 (#204)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Expert | 1266 | 1259 |
| GPQA Diamond | — | 67% |
| Humanity's Last Exam | — | 5.7% |
| MMLU-Pro | — | 81% |
| Confabulations | — | 22.6% |
| Vectara Hallucination Rate | — | 8.2% |
| GPQA (HELM) | — | 65% |
Multimodal Not comparable
DeepSeek-V2.5 (Sep 2024): —, Llama 4 Maverick: 31.6 (#105)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Vision | — | 1142 |
| GeoBench | — | 52% |
| SpatialViz-Bench | — | 31.8% |
Multilingual Too close to call
DeepSeek-V2.5 (Sep 2024): 42.5 (#193), Llama 4 Maverick: 42.2 (#195)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Non-English | 1273 | 1269 |
| LMArena Chinese | 1318 | 1277 |
| LMArena French | 1289 | 1259 |
| LMArena German | 1258 | 1291 |
| LMArena Japanese | 1228 | 1207 |
| LMArena Korean | 1209 | 1203 |
| LMArena Russian | 1289 | 1286 |
| LMArena Spanish | 1248 | 1293 |
Instruction Following Llama 4 Maverick leads
DeepSeek-V2.5 (Sep 2024): 67.5 (#194), Llama 4 Maverick: 71.7 (#146)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Instruction Following | 1280 | 1267 |
| IFEval | — | 90.8% |
Long Context DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 39.5 (#174), Llama 4 Maverick: 31.4 (#279)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Longer Query | 1301 | 1280 |
| Fiction.LiveBench | — | 46.2% |
Writing & Preference DeepSeek-V2.5 (Sep 2024) leads
DeepSeek-V2.5 (Sep 2024): 49.8 (#187), Llama 4 Maverick: 38.8 (#252)
| Benchmark | DeepSeek-V2.5 (Sep 2024) | Llama 4 Maverick |
|---|---|---|
| LMArena Text | 1294 | 1287 |
| LMArena Creative Writing | 1285 | 1267 |
| LMArena Multi-Turn | 1297 | 1289 |
| Short-Story Creative Writing | — | 62% |
| EQ-Bench Creative Writing | — | 860 |
| WildBench | — | 80% |
Frequently asked questions
Is DeepSeek-V2.5 (Sep 2024) better than Llama 4 Maverick?
DeepSeek-V2.5 (Sep 2024) is the stronger model overall, scoring 37.6 to 30.9 on the Noometry Index.
Is DeepSeek-V2.5 (Sep 2024) or Llama 4 Maverick better for coding?
DeepSeek-V2.5 (Sep 2024) scores higher on coding benchmarks: 31.7 versus 26.6 in the Noometry coding category.
How many benchmarks do DeepSeek-V2.5 (Sep 2024) and Llama 4 Maverick share?
20 benchmarks have published results for both models. DeepSeek-V2.5 (Sep 2024) has 22 scored results on Noometry and Llama 4 Maverick has 54.