Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
Qwen3.6 Plus
69
Winner · 2/8 categoriesSarvam 30B
48
1/8 categoriesQwen3.6 Plus· Sarvam 30B
Pick Qwen3.6 Plus if you want the stronger benchmark profile. Sarvam 30B only becomes the better choice if knowledge is the priority.
Qwen3.6 Plus is clearly ahead on the aggregate, 69 to 48. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Qwen3.6 Plus's sharpest advantage is in coding, where it averages 64.9 against 34. The single biggest benchmark swing on the page is SWE-bench Verified, 78.8% to 34%. Sarvam 30B does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.
Qwen3.6 Plus gives you the larger context window at 1M, 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 | Qwen3.6 Plus | Sarvam 30B |
|---|---|---|
| AgenticQwen3.6 Plus wins | ||
| Terminal-Bench 2.0 | 61.6% | — |
| Claw-Eval | 58.7% | — |
| QwenClawBench | 57.2% | — |
| QwenWebBench | 1502 | — |
| TAU3-Bench | 70.7% | — |
| VITA-Bench | 44.3% | — |
| DeepPlanning | 41.5% | — |
| Toolathlon | 39.8% | — |
| MCP Atlas | 48.2% | — |
| MCP-Tasks | 74.1% | — |
| WideResearch | 74.3% | — |
| OSWorld-Verified | 62.5% | — |
| BrowseComp | — | 35.5% |
| CodingQwen3.6 Plus wins | ||
| SWE-bench Verified | 78.8% | 34% |
| SWE-bench Pro | 56.6% | — |
| SWE Multilingual | 73.8% | — |
| LiveCodeBench v6 | 87.1% | 70.0% |
| NL2Repo | 37.9% | — |
| HumanEval | — | 92.1% |
| Multimodal & Grounded | ||
| MMMU | 86.0% | — |
| MMMU-Pro | 78.8% | — |
| RealWorldQA | 85.4% | — |
| OmniDocBench 1.5 | 91.2% | — |
| Video-MME (with subtitle) | 87.8% | — |
| Video-MME (w/o subtitle) | 84.2% | — |
| MathVision | 88.0% | — |
| We-Math | 89.0% | — |
| DynaMath | 88.0% | — |
| MStar | 83.3% | — |
| SimpleVQA | 67.3% | — |
| ChatCVQA | 81.5% | — |
| MMLongBench-Doc | 62.0% | — |
| CC-OCR | 83.4% | — |
| AI2D_TEST | 94.4% | — |
| CountBench | 97.6% | — |
| RefCOCO (avg) | 93.5% | — |
| ODINW13 | 51.8% | — |
| ERQA | 65.7% | — |
| VideoMMMU | 84.0% | — |
| MLVU (M-Avg) | 86.7% | — |
| ScreenSpot Pro | 68.2% | — |
| Reasoning | ||
| AI-Needle | 68.3% | — |
| LongBench v2 | 62% | — |
| gpqaDiamond | — | 66.5% |
| KnowledgeSarvam 30B wins | ||
| GPQA | 90.4% | — |
| SuperGPQA | 71.6% | — |
| MMLU-Pro | 88.5% | 80% |
| MMLU-Redux | 94.5% | — |
| C-Eval | 93.3% | — |
| HLE | 28.8% | — |
| MMLU | — | 85.1% |
| Instruction Following | ||
| IFEval | 94.3% | — |
| IFBench | 74.2% | — |
| Multilingual | ||
| MMLU-ProX | 84.7% | — |
| NOVA-63 | 57.9% | — |
| INCLUDE | 85.1% | — |
| PolyMath | 77.4% | — |
| VWT2k-lite | 84.3% | — |
| MAXIFE | 88.2% | — |
| Mathematics | ||
| AIME26 | 95.3% | — |
| HMMT Feb 2025 | 96.7% | 73.3% |
| HMMT Nov 2025 | 94.6% | 74.2% |
| HMMT Feb 2026 | 87.8% | — |
| MMAnswerBench | 83.8% | — |
| MATH-500 | — | 97% |
| AIME 2025 | — | 80% |
Qwen3.6 Plus is ahead overall, 69 to 48. The biggest single separator in this matchup is SWE-bench Verified, where the scores are 78.8% and 34%.
Sarvam 30B has the edge for knowledge tasks in this comparison, averaging 80 versus 66. Inside this category, MMLU-Pro is the benchmark that creates the most daylight between them.
Qwen3.6 Plus has the edge for coding in this comparison, averaging 64.9 versus 34. Inside this category, SWE-bench Verified is the benchmark that creates the most daylight between them.
Qwen3.6 Plus has the edge for agentic tasks in this comparison, averaging 62 versus 35.5. Sarvam 30B stays close enough that the answer can still flip depending on your workload.
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.