Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
Qwen3.5 397B
68
Winner · 3/8 categoriesSarvam 105B
60
2/8 categoriesQwen3.5 397B· Sarvam 105B
Pick Qwen3.5 397B if you want the stronger benchmark profile. Sarvam 105B only becomes the better choice if knowledge is the priority or you want the stronger reasoning-first profile.
Qwen3.5 397B is clearly ahead on the aggregate, 68 to 60. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Qwen3.5 397B's sharpest advantage is in agentic, where it averages 58.3 against 49.5. The single biggest benchmark swing on the page is SWE-bench Verified, 76.2% to 45%. Sarvam 105B does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.
Sarvam 105B is the reasoning model in the pair, while Qwen3.5 397B 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.
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 | Qwen3.5 397B | Sarvam 105B |
|---|---|---|
| AgenticQwen3.5 397B wins | ||
| Terminal-Bench 2.0 | 52.5% | — |
| BrowseComp | 62% | 49.5% |
| OSWorld-Verified | 62.2% | — |
| Claw-Eval | 48.1% | — |
| QwenClawBench | 51.8% | — |
| QwenWebBench | 1162 | — |
| TAU3-Bench | 68.4% | — |
| DeepPlanning | 37.6% | — |
| Toolathlon | 36.3% | — |
| MCP Atlas | 46.1% | — |
| MCP-Tasks | 74.2% | — |
| WideResearch | 74.0% | — |
| CodingQwen3.5 397B wins | ||
| HumanEval | 75% | — |
| SWE-bench Verified | 76.2% | 45% |
| LiveCodeBench | 39% | — |
| LiveCodeBench v6 | 83.6% | 71.7% |
| SWE-bench Pro | 50.9% | — |
| SWE Multilingual | 69.3% | — |
| NL2Repo | 32.2% | — |
| Multimodal & Grounded | ||
| MMMU-Pro | 79% | — |
| OfficeQA Pro | 68% | — |
| RealWorldQA | 83.9% | — |
| Video-MME (w/o subtitle) | 84.2% | — |
| MathVision | 88.6% | — |
| We-Math | 87.9% | — |
| DynaMath | 86.3% | — |
| MStar | 83.8% | — |
| SimpleVQA | 67.1% | — |
| ChatCVQA | 80.8% | — |
| AI2D_TEST | 93.9% | — |
| CountBench | 97.2% | — |
| RefCOCO (avg) | 92.3% | — |
| ODINW13 | 47.0% | — |
| MLVU (M-Avg) | 86.7% | — |
| ScreenSpot Pro | 65.6% | — |
| Reasoning | ||
| MuSR | 78% | — |
| BBH | 82% | — |
| LongBench v2 | 63.2% | — |
| MRCRv2 | 71% | — |
| AI-Needle | 68.7% | — |
| gpqaDiamond | — | 78.7% |
| hle | — | 11.2% |
| KnowledgeSarvam 105B wins | ||
| MMLU | 83% | 90.6% |
| GPQA | 88.4% | — |
| SuperGPQA | 70.4% | — |
| MMLU-Pro | 87.8% | 81.7% |
| MMLU-Redux | 94.9% | — |
| C-Eval | 93% | — |
| HLE | 28.7% | — |
| FrontierScience | 71% | — |
| SimpleQA | 80% | — |
| Instruction FollowingQwen3.5 397B wins | ||
| IFEval | 92.6% | 84.8% |
| IFBench | 76.5% | — |
| Multilingual | ||
| MGSM | 82% | — |
| MMLU-ProX | 84.7% | — |
| NOVA-63 | 59.1% | — |
| INCLUDE | 85.6% | — |
| PolyMath | 73.3% | — |
| VWT2k-lite | 78.9% | — |
| MAXIFE | 88.2% | — |
| MathematicsSarvam 105B wins | ||
| AIME 2023 | 83% | — |
| AIME 2024 | 85% | — |
| AIME 2025 | 84% | 88.3% |
| AIME26 | 93.3% | — |
| HMMT Feb 2023 | 79% | — |
| HMMT Feb 2024 | 81% | — |
| HMMT Feb 2025 | 80% | — |
| HMMT Feb 2025 | 94.8% | 85.8% |
| HMMT Nov 2025 | 92.7% | 85.8% |
| HMMT Feb 2026 | 87.9% | — |
| MMAnswerBench | 80.9% | — |
| BRUMO 2025 | 82% | — |
| MATH-500 | 81% | 98.6% |
Qwen3.5 397B is ahead overall, 68 to 60. The biggest single separator in this matchup is SWE-bench Verified, where the scores are 76.2% and 45%.
Sarvam 105B has the edge for knowledge tasks in this comparison, averaging 81.7 versus 68.2. Inside this category, MMLU is the benchmark that creates the most daylight between them.
Qwen3.5 397B has the edge for coding in this comparison, averaging 52.2 versus 45. Inside this category, SWE-bench Verified is the benchmark that creates the most daylight between them.
Sarvam 105B has the edge for math in this comparison, averaging 92.3 versus 82.6. Inside this category, MATH-500 is the benchmark that creates the most daylight between them.
Qwen3.5 397B has the edge for agentic tasks in this comparison, averaging 58.3 versus 49.5. Inside this category, BrowseComp is the benchmark that creates the most daylight between them.
Qwen3.5 397B has the edge for instruction following in this comparison, averaging 92.6 versus 84.8. Inside this category, IFEval 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.