Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
Kimi K2.5
72
Winner · 2/8 categoriesSarvam 30B
48
2/8 categoriesKimi K2.5· Sarvam 30B
Pick Kimi K2.5 if you want the stronger benchmark profile. Sarvam 30B only becomes the better choice if knowledge is the priority or you want the cheaper token bill.
Kimi K2.5 is clearly ahead on the aggregate, 72 to 48. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
Kimi K2.5's sharpest advantage is in coding, where it averages 66.7 against 34. The single biggest benchmark swing on the page is SWE-bench Verified, 76.8% 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.
Kimi K2.5 is also the more expensive model on tokens at $0.50 input / $2.80 output per 1M tokens, versus $0.00 input / $0.00 output per 1M tokens for Sarvam 30B. That is roughly Infinityx on output cost alone. Sarvam 30B is the reasoning model in the pair, while Kimi K2.5 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. Kimi K2.5 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 | Kimi K2.5 | Sarvam 30B |
|---|---|---|
| AgenticKimi K2.5 wins | ||
| Terminal-Bench 2.0 | 50.8% | — |
| BrowseComp | 60.6% | 35.5% |
| OSWorld-Verified | 63.3% | — |
| Claw-Eval | 52.9% | — |
| QwenClawBench | 54.3% | — |
| QwenWebBench | 1160 | — |
| TAU3-Bench | 65.7% | — |
| VITA-Bench | 36.0% | — |
| DeepPlanning | 14.4% | — |
| Toolathlon | 27.8% | — |
| MCP Atlas | 29.5% | — |
| MCP-Tasks | 59.1% | — |
| WideResearch | 72.7% | — |
| BrowseComp-VL | 42.9% | — |
| OSWorld | 63.3% | — |
| Tau2-Airline | 80.0% | — |
| Tau2-Telecom | 95.9% | — |
| BFCL v4 | 68.3% | — |
| AndroidWorld | 43.1% | — |
| WebVoyager | 84.3% | — |
| CodingKimi K2.5 wins | ||
| HumanEval | 99% | 92.1% |
| SWE-bench Verified | 76.8% | 34% |
| SWE-bench Verified* | 70.8% | — |
| LiveCodeBench | 85% | — |
| LiveCodeBench v6 | 85.6% | 70.0% |
| SWE-bench Pro | 53.8% | — |
| SWE Multilingual | 73% | — |
| SWE-Rebench | 58.5% | — |
| React Native Evals | 74.9% | — |
| NL2Repo | 32% | — |
| Multimodal & Grounded | ||
| MMMU | 84.3% | — |
| MMMU-Pro | 78.5% | — |
| OfficeQA Pro | 69% | — |
| RealWorldQA | 81.0% | — |
| OmniDocBench 1.5 | 88.8% | — |
| Video-MME (with subtitle) | 87.4% | — |
| Video-MME (w/o subtitle) | 83.2% | — |
| MathVision | 84.2% | — |
| We-Math | 84.7% | — |
| DynaMath | 84.4% | — |
| MStar | 80.5% | — |
| Design2Code | 91.3% | — |
| Flame-VLM-Code | 88.8% | — |
| Vision2Web | 33.2% | — |
| ImageMining | 24.4% | — |
| MMSearch | 58.7% | — |
| MMSearch-Plus | 25.6% | — |
| SimpleVQA | 71.2% | — |
| ChatCVQA | 77.5% | — |
| MMLongBench-Doc | 58.5% | — |
| CC-OCR | 79.7% | — |
| AI2D_TEST | 90.8% | — |
| CountBench | 94.1% | — |
| RefCOCO (avg) | 87.8% | — |
| VideoMMMU | 86.6% | — |
| MLVU (M-Avg) | 85.0% | — |
| Facts-VLM | 57.8% | — |
| V* | 84.3% | — |
| Reasoning | ||
| MuSR | 72% | — |
| BBH | 81% | — |
| LongBench v2 | 61% | — |
| MRCRv2 | 70% | — |
| AI-Needle | 70% | — |
| gpqaDiamond | — | 66.5% |
| KnowledgeSarvam 30B wins | ||
| MMLU | 77% | 85.1% |
| GPQA | 87.6% | — |
| GPQA-D | 86.9% | — |
| SuperGPQA | 69.2% | — |
| MMLU-Pro | 87.1% | 80% |
| MMLU-Pro (Arcee) | 87.1% | — |
| MMLU-Redux | 94.5% | — |
| C-Eval | 94% | — |
| HLE | 30.1% | — |
| FrontierScience | 67% | — |
| SimpleQA | 74% | — |
| Instruction Following | ||
| IFEval | 93.9% | — |
| IFBench | 70.2% | — |
| Multilingual | ||
| MGSM | 83% | — |
| MMLU-ProX | 82.3% | — |
| NOVA-63 | 56.0% | — |
| INCLUDE | 83.3% | — |
| PolyMath | 43.1% | — |
| VWT2k-lite | 77.6% | — |
| MAXIFE | 72.8% | — |
| MathematicsSarvam 30B wins | ||
| AIME 2023 | 77% | — |
| AIME 2024 | 79% | — |
| AIME 2025 | 78% | 80% |
| AIME26 | 95.8% | — |
| AIME25 (Arcee) | 96.3% | — |
| HMMT Feb 2023 | 73% | — |
| HMMT Feb 2024 | 75% | — |
| HMMT Feb 2025 | 74% | — |
| HMMT Feb 2025 | 95.4% | 73.3% |
| HMMT Nov 2025 | 91.1% | 74.2% |
| HMMT Feb 2026 | 87.1% | — |
| MMAnswerBench | 81.8% | — |
| BRUMO 2025 | 76% | — |
| MATH-500 | 82% | 97% |
Kimi K2.5 is ahead overall, 72 to 48. The biggest single separator in this matchup is SWE-bench Verified, where the scores are 76.8% and 34%.
Sarvam 30B has the edge for knowledge tasks in this comparison, averaging 80 versus 66.6. Inside this category, MMLU is the benchmark that creates the most daylight between them.
Kimi K2.5 has the edge for coding in this comparison, averaging 66.7 versus 34. Inside this category, SWE-bench Verified is the benchmark that creates the most daylight between them.
Sarvam 30B has the edge for math in this comparison, averaging 86.5 versus 78.3. Inside this category, HMMT Feb 2025 is the benchmark that creates the most daylight between them.
Kimi K2.5 has the edge for agentic tasks in this comparison, averaging 57.6 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.