Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
DeepSeek V3.2
61
Winner · 2/8 categoriesSarvam 30B
48
2/8 categoriesDeepSeek V3.2· Sarvam 30B
Pick DeepSeek V3.2 if you want the stronger benchmark profile. Sarvam 30B only becomes the better choice if knowledge is the priority or you want the stronger reasoning-first profile.
DeepSeek V3.2 is clearly ahead on the aggregate, 61 to 48. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
DeepSeek V3.2's sharpest advantage is in agentic, where it averages 58.8 against 35.5. The single biggest benchmark swing on the page is BrowseComp, 62% to 35.5%. Sarvam 30B does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.
Sarvam 30B is the reasoning model in the pair, while DeepSeek V3.2 is not. That usually helps on harder chain-of-thought-heavy tests, but it can also mean more latency and more token spend in real use. DeepSeek V3.2 gives you the larger context window at 128K, compared with 64K for Sarvam 30B.
BenchLM keeps the benchmark table and the operator tradeoffs on the same page so a better score does not hide a materially slower, pricier, or smaller-context model.
Runtime metrics show N/A when BenchLM does not have a sourced snapshot for that exact model. The scoring rules and freshness policy are documented on the methodology page.
| Benchmark | DeepSeek V3.2 | Sarvam 30B |
|---|---|---|
| AgenticDeepSeek V3.2 wins | ||
| Terminal-Bench 2.0 | 60% | — |
| BrowseComp | 62% | 35.5% |
| OSWorld-Verified | 55% | — |
| Claw-Eval | 51.0% | — |
| DeepPlanning | 19.0% | — |
| VITA-Bench | 18.5% | — |
| CodingDeepSeek V3.2 wins | ||
| HumanEval | 76% | 92.1% |
| SWE-bench Verified | 45% | 34% |
| SWE-Rebench | 60.9% | — |
| React Native Evals | 69% | — |
| LiveCodeBench v6 | — | 70.0% |
| Multimodal & Grounded | ||
| MMMU-Pro | 61% | — |
| OfficeQA Pro | 72% | — |
| Reasoning | ||
| LongBench v2 | 69% | — |
| MRCRv2 | 70% | — |
| ARC-AGI-2 | 4% | — |
| gpqaDiamond | — | 66.5% |
| KnowledgeSarvam 30B wins | ||
| GPQA | 83% | — |
| HLE | 11% | — |
| FrontierScience | 72% | — |
| MMLU | — | 85.1% |
| MMLU-Pro | — | 80% |
| Instruction Following | ||
| IFEval | 85% | — |
| Multilingual | ||
| MGSM | 84% | — |
| MMLU-ProX | 81% | — |
| MathematicsSarvam 30B wins | ||
| AIME 2023 | 84% | — |
| AIME 2024 | 86% | — |
| AIME 2025 | 85% | 80% |
| HMMT Feb 2023 | 80% | — |
| MATH-500 | 81% | 97% |
| HMMT Feb 2025 | — | 73.3% |
| HMMT Nov 2025 | — | 74.2% |
DeepSeek V3.2 is ahead overall, 61 to 48. The biggest single separator in this matchup is BrowseComp, where the scores are 62% and 35.5%.
Sarvam 30B has the edge for knowledge tasks in this comparison, averaging 80 versus 48. DeepSeek V3.2 stays close enough that the answer can still flip depending on your workload.
DeepSeek V3.2 has the edge for coding in this comparison, averaging 56.1 versus 34. Inside this category, HumanEval is the benchmark that creates the most daylight between them.
Sarvam 30B has the edge for math in this comparison, averaging 86.5 versus 83.5. Inside this category, MATH-500 is the benchmark that creates the most daylight between them.
DeepSeek V3.2 has the edge for agentic tasks in this comparison, averaging 58.8 versus 35.5. Inside this category, BrowseComp is the benchmark that creates the most daylight between them.
Get notified when new models drop, benchmark scores change, or the leaderboard shifts. One email per week.
Free. No spam. Unsubscribe anytime. We only store derived location metadata for consent routing.