Side-by-side benchmark comparison across agentic, coding, multimodal, knowledge, reasoning, and math workflows.
GPT-5.1-Codex-Max
81
Winner · 4/8 categoriesSarvam 105B
60
1/8 categoriesGPT-5.1-Codex-Max· Sarvam 105B
Pick GPT-5.1-Codex-Max if you want the stronger benchmark profile. Sarvam 105B only becomes the better choice if knowledge is the priority or you want the cheaper token bill.
GPT-5.1-Codex-Max is clearly ahead on the aggregate, 81 to 60. The gap is large enough that you do not need to squint at the spreadsheet to see the difference.
GPT-5.1-Codex-Max's sharpest advantage is in agentic, where it averages 86 against 49.5. The single biggest benchmark swing on the page is BrowseComp, 85% to 49.5%. Sarvam 105B does hit back in knowledge, so the answer changes if that is the part of the workload you care about most.
GPT-5.1-Codex-Max is also the more expensive model on tokens at $2.00 input / $8.00 output per 1M tokens, versus $0.00 input / $0.00 output per 1M tokens for Sarvam 105B. That is roughly Infinityx on output cost alone. GPT-5.1-Codex-Max gives you the larger context window at 400K, compared with 128K for Sarvam 105B.
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 | GPT-5.1-Codex-Max | Sarvam 105B |
|---|---|---|
| AgenticGPT-5.1-Codex-Max wins | ||
| Terminal-Bench 2.0 | 90% | — |
| BrowseComp | 85% | 49.5% |
| OSWorld-Verified | 82% | — |
| CodingGPT-5.1-Codex-Max wins | ||
| HumanEval | 94% | — |
| SWE-bench Verified | 77.9% | 45% |
| LiveCodeBench | 67% | — |
| SWE-bench Pro | 84% | — |
| LiveCodeBench v6 | — | 71.7% |
| Multimodal & Grounded | ||
| MMMU-Pro | 85% | — |
| OfficeQA Pro | 92% | — |
| Reasoning | ||
| MuSR | 92% | — |
| BBH | 92% | — |
| LongBench v2 | 90% | — |
| MRCRv2 | 93% | — |
| gpqaDiamond | — | 78.7% |
| hle | — | 11.2% |
| KnowledgeSarvam 105B wins | ||
| MMLU | 98% | 90.6% |
| GPQA | 96% | — |
| SuperGPQA | 94% | — |
| MMLU-Pro | 82% | 81.7% |
| HLE | 27% | — |
| FrontierScience | 84% | — |
| SimpleQA | 94% | — |
| Instruction FollowingGPT-5.1-Codex-Max wins | ||
| IFEval | 91% | 84.8% |
| Multilingual | ||
| MGSM | 89% | — |
| MMLU-ProX | 87% | — |
| MathematicsGPT-5.1-Codex-Max wins | ||
| AIME 2023 | 99% | — |
| AIME 2024 | 99% | — |
| AIME 2025 | 98% | 88.3% |
| HMMT Feb 2023 | 95% | — |
| HMMT Feb 2024 | 97% | — |
| HMMT Feb 2025 | 96% | — |
| BRUMO 2025 | 96% | — |
| MATH-500 | 93% | 98.6% |
| HMMT Feb 2025 | — | 85.8% |
| HMMT Nov 2025 | — | 85.8% |
GPT-5.1-Codex-Max is ahead overall, 81 to 60. The biggest single separator in this matchup is BrowseComp, where the scores are 85% and 49.5%.
Sarvam 105B has the edge for knowledge tasks in this comparison, averaging 81.7 versus 74.4. Inside this category, MMLU is the benchmark that creates the most daylight between them.
GPT-5.1-Codex-Max has the edge for coding in this comparison, averaging 76.1 versus 45. Inside this category, SWE-bench Verified is the benchmark that creates the most daylight between them.
GPT-5.1-Codex-Max has the edge for math in this comparison, averaging 96.1 versus 92.3. Inside this category, AIME 2025 is the benchmark that creates the most daylight between them.
GPT-5.1-Codex-Max has the edge for agentic tasks in this comparison, averaging 86 versus 49.5. Inside this category, BrowseComp is the benchmark that creates the most daylight between them.
GPT-5.1-Codex-Max has the edge for instruction following in this comparison, averaging 91 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.