Model comparison
DeepSeek-V3.2-Speciale vs Llama 4 Maverick
DeepSeek-V3.2-Speciale is the stronger model overall, scoring 39.1 to 30.6 on the Noometry Index. Llama 4 Maverick costs 2.8× less per token, which makes it the better buy when DeepSeek-V3.2-Speciale's lead doesn't matter for your workload.
Last verified . 3 shared benchmarks.
DeepSeek
MetaSummary
- They share 3 benchmarks with published results for both. DeepSeek-V3.2-Speciale scores higher in 3 categories and Llama 4 Maverick in 0 categories; 3 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where DeepSeek-V3.2-Speciale leads 32.1 to 10.1.
- The biggest single-benchmark swing is SimpleBench: 52.6% for DeepSeek-V3.2-Speciale and 27.7% for Llama 4 Maverick.
- Llama 4 Maverick is cheaper at $0.19 / $0.65 per million input/output tokens, against $0.58 / $1.68 for DeepSeek-V3.2-Speciale.
Side by side
| DeepSeek-V3.2-Speciale | Llama 4 Maverick | |
|---|---|---|
| Provider | DeepSeek | Meta |
| Noometry Index | 39.1 | 30.6 |
| Released | 2025-12-01 | 2025-04-05 |
| Weights | Open | Open |
| Context window | 128K | 128K |
| Max output | 128K | 4K |
| Input $ / M tokens | $0.58 | $0.19 |
| Output $ / M tokens | $1.68 | $0.65 |
| Results tracked | 3 | 54 |
Sponsored placements are available on pages like this one. Advertise on Noometry
Category by category
Coding DeepSeek-V3.2-Speciale leads
DeepSeek-V3.2-Speciale: 39.9 (#136), Llama 4 Maverick: 26.3 (#318)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| WeirdML | 46.7% | 24.5% |
| SWE-bench Verified (bash only) | — | 21% |
| Aider Polyglot | — | 15.6% |
| SciCode | — | 33.1% |
| BigCodeBench Instruct | — | 49.7% |
| LMArena Coding | — | 1302 |
| BigCodeBench Complete | — | 61.4% |
| ALE-Bench | — | 172.97 |
Agentic & Tool Use Not comparable
DeepSeek-V3.2-Speciale: —, Llama 4 Maverick: 27.9 (#86)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| Berkeley Function Calling Leaderboard | — | 37.3% |
Reasoning DeepSeek-V3.2-Speciale leads
DeepSeek-V3.2-Speciale: 32.1 (#65), Llama 4 Maverick: 10.1 (#337)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| SimpleBench | 52.6% | 27.7% |
| ARC-AGI-2 | — | 0% |
| Kagi LLM Benchmark | — | 55.9% |
| NYT Connections (extended) | — | 8% |
| ARC-AGI-1 | — | 4.4% |
| CritPt | — | 0% |
| EnigmaEval | — | 0.6% |
| LMArena Hard Prompts | — | 1281 |
| DTBench | — | 61.9% |
| LMCA | — | 15.9% |
| Epoch Capabilities Index | — | 132.2 |
| ForecastBench | — | 57.5 |
Math Not comparable
DeepSeek-V3.2-Speciale: —, Llama 4 Maverick: 24.6 (#256)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| OTIS Mock AIME 2024-2025 | — | 20.6% |
| Omni-MATH | — | 42.2% |
| LMArena Math | — | 1299 |
| MATH Level 5 | — | 73% |
| FrontierMath (Feb 2025 set) | — | 0.7% |
Knowledge Not comparable
DeepSeek-V3.2-Speciale: —, Llama 4 Maverick: 33.3 (#196)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| GPQA Diamond | — | 67% |
| Humanity's Last Exam | — | 5.7% |
| MMLU-Pro | — | 81% |
| Confabulations | — | 22.6% |
| Vectara Hallucination Rate | — | 8.2% |
| GPQA (HELM) | — | 65% |
| LMArena Expert | — | 1259 |
Multimodal Not comparable
DeepSeek-V3.2-Speciale: —, Llama 4 Maverick: 30.9 (#100)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| LMArena Vision | — | 1141 |
| GeoBench | — | 52% |
| SpatialViz-Bench | — | 31.8% |
Multilingual Not comparable
DeepSeek-V3.2-Speciale: —, Llama 4 Maverick: 42.4 (#190)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| LMArena Non-English | — | 1269 |
| LMArena Chinese | — | 1277 |
| LMArena French | — | 1258 |
| LMArena German | — | 1291 |
| LMArena Japanese | — | 1207 |
| LMArena Korean | — | 1203 |
| LMArena Russian | — | 1286 |
| LMArena Spanish | — | 1293 |
Instruction Following Not comparable
DeepSeek-V3.2-Speciale: —, Llama 4 Maverick: 71.9 (#140)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| IFEval | — | 90.8% |
| LMArena Instruction Following | — | 1267 |
Long Context Not comparable
DeepSeek-V3.2-Speciale: —, Llama 4 Maverick: 31.4 (#274)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| Fiction.LiveBench | — | 46.2% |
| LMArena Longer Query | — | 1280 |
Writing & Preference DeepSeek-V3.2-Speciale leads
DeepSeek-V3.2-Speciale: 46.1 (#217), Llama 4 Maverick: 39.0 (#247)
| Benchmark | DeepSeek-V3.2-Speciale | Llama 4 Maverick |
|---|---|---|
| EQ-Bench Creative Writing | 1276 | 860 |
| LMArena Text | — | 1287 |
| LMArena Creative Writing | — | 1266 |
| Short-Story Creative Writing | — | 62% |
| WildBench | — | 80% |
| LMArena Multi-Turn | — | 1289 |
Frequently asked questions
Is DeepSeek-V3.2-Speciale better than Llama 4 Maverick?
DeepSeek-V3.2-Speciale is the stronger model overall, scoring 39.1 to 30.6 on the Noometry Index. Llama 4 Maverick costs 2.8× less per token, which makes it the better buy when DeepSeek-V3.2-Speciale's lead doesn't matter for your workload.
Which is cheaper, DeepSeek-V3.2-Speciale or Llama 4 Maverick?
Llama 4 Maverick is cheaper. It lists at $0.19 per million input tokens and $0.65 per million output tokens; DeepSeek-V3.2-Speciale lists at $0.58 and $1.68.
Is DeepSeek-V3.2-Speciale or Llama 4 Maverick better for coding?
DeepSeek-V3.2-Speciale scores higher on coding benchmarks: 39.9 versus 26.3 in the Noometry coding category.
Which has the bigger context window?
Both accept 128K tokens.
How many benchmarks do DeepSeek-V3.2-Speciale and Llama 4 Maverick share?
3 benchmarks have published results for both models. DeepSeek-V3.2-Speciale has 3 scored results on Noometry and Llama 4 Maverick has 54.