Sage-S2.1-Flash vs Maxwell-V1.0-Flash
New ComparisonScores are fetched live from the current database on every visit — nothing is cached.
Winner
Sage-S2.1-Flash wins this comparison with an overall score of 88 vs 49 for Maxwell-V1.0-Flash, leading 4 of 4 shared benchmarks.
Based on 4 of 83 benchmarks scored by all 2 selected models.
Benchmark Scores Comparison
Sage-S2.1-Flash
321.3B params
Maxwell-V1.0-Flash
37.0B params
Coverage matrix
2 models × 83 benchmarks — hover a benchmark name for its category
| Benchmark | Maxwell-V1.0-Flash Maxwell | Sage-S2.1-Flash Shanghai AI Laboratory |
|---|---|---|
| OSWorld-G | — | 84.8 |
| HMMT Feb 25 | 100 | — |
| GPQA Diamond | 86.52 | — |
| MGSM | 93 | — |
| LiveCodeBench v6 (pass@8) | 93.74 | — |
| HLE (text-only) | 24.79 | — |
| LiveCodeBench v6 | 85.66 | — |
| Automation-Bench | — | 56.7 |
| SWE-bench Pro | — | 60.6 |
| PopQA-300 | 70.1 | — |
| BFCL-V4 | — | 75.5 |
| AIME 2025 (mean@8) | 96.25 | — |
| HMMT Feb 25 (mean@8) | 96.25 | — |
| AA-LCR | 75.6 | — |
| AIME 26 | 96.67 | — |
| HMMT Feb 26 (mean@8) | 91.29 | — |
| TAU3-Bench | 67.37 | — |
| 2WikiMultihopQA-300 | 89.8 | — |
| GDPVal | — | 78 |
| OfficeQA Pro | — | 72.2 |
| DeepSearch QA | 83.3 | 94.4 |
| RoleBench | 89.92 | — |
| MLE-bench Lite | — | 90.3 |
| HMMT Nov 25 (mean@8) | 90.83 | — |
| MLS-Bench | — | 42 |
| PostTrainBench | — | 39.2 |
| INCLUDE | 84.7 | — |
| BabyVision | — | 75.8 |
| MMMLU | 92 | — |
| OSWorld V2.1 | — | 69.3 |
| HMMT Nov 25 | 100 | — |
| BrowseComp-VL | — | 60.5 |
| Apex-Agents | — | 61.7 |
| Multi-Challenge | 72.2 | — |
| WorkSpaceBench | — | 61.7 |
| MMLU-Pro | 85.46 | — |
| OSWorld-Verified | — | 85.2 |
| HotpotQA-300 | 84.7 | — |
| Global PIQA | 92.3 | — |
| DeepresearchBII | — | 54.2 |
| Workspace-Bench-Lite | — | 67.1 |
| SpreadsheetBench 2 | — | 37.1 |
| IFEval | 93 | — |
| HLE (with tools) | 46.1 | — |
| MLogiQA | 81.5 | — |
| Belebele (5-language subset) | 92.8 | — |
| KOR-Bench | 77.2 | — |
| WebWalkerQA-300 | 70.5 | — |
| BrowseComp | 76.3 | 91.2 |
| MMLU-Redux | 93.59 | — |
| NL2Repo-Bench | — | 57.2 |
| SkillsBench | — | 62.3 |
| WebArena-Verified | — | 67.6 |
| CoSER | 53.54 | — |
| MCPMark | 37.3 | — |
| Harvey Legal Agent | — | 25 |
| PaperBench | — | 88 |
| Agents' Last Exam (CLI) | — | 32.4 |
| IFBench | 64.8 | — |
| SWE-bench Verified | 65.4 | — |
| LongMemEval | 86.18 | — |
| IMOAnswerBench | 95.5 | — |
| BIG-Bench Hard | 91.02 | — |
| AIME 26 (mean@8) | 95 | — |
| MuSiQue-300 | 63 | — |
| AdvancedIF | 71.3 | — |
| MT-RAG | 65.74 | — |
| MT-Bench-101 | 93.3 | — |
| Terminal-Bench 2.1 (Terminus-2) | 58.65 | 87.9 |
| DeepSWE | — | 67.9 |
| HMMT Feb 26 | 100 | — |
| AIME 2025 | 100 | — |
| IMO-AnswerBench (mean@8) | 86.03 | — |
| JobBench | — | 58.8 |
| RULER (32K context) | 95.81 | — |
| Bamboogle | 92.1 | — |
| C-Eval | 91 | — |
| WideSearch | 74.4 | 81.5 |
| Vision2Web | — | 78 |
| TauBench V3 Banking | — | 45.4 |
| LongBench v2 | 59.4 | — |
| BigBench Extra Hard | 52.51 | — |
| WeaveBench | — | 68.4 |
| Covered | 54/83 | 33/83 |
Scorecard
Overall 0–100 score — category strengths (50%), shared-benchmark wins (30%), coverage (20%)
| Model | Overall score | Shared benchmark wins | Coverage | Strongest at | Weakest at |
|---|---|---|---|---|---|
| Sage-S2.1-Flash Winner | 88 | 4 of 4 | 33 of 83 (40.0%) | General Agents | Agentic Coding |
| Maxwell-V1.0-Flash | 49.3 | 0 of 4 | 54 of 83 (65.0%) | General Agents | Agentic Coding |
Category strengths
Where each model stands, per benchmark category — score relative to the best model, adjusted for benchmark coverage. This also ranks models when they share few benchmarks.
| Category | Sage-S2.1-Flash | Maxwell-V1.0-Flash |
|---|---|---|
| General Agents 30 benchmarks | 100 | 77.9 |
| Agentic Coding 9 benchmarks | 100 | 56.3 |
Fully comparable benchmarks
Scored by all 2 selected models (4)
| Benchmark | Category | Sage-S2.1-Flash | Maxwell-V1.0-Flash |
|---|---|---|---|
| DeepSearch QA | general_agent | 94.4 | 83.3 |
| BrowseComp | general_agent | 91.2 | 76.3 |
| Terminal-Bench 2.1 (Terminus-2) | coding_agent | 87.9 | 58.65 |
| WideSearch | general_agent | 81.5 | 74.4 |
Partial coverage benchmarks
Scored by at least one but not all selected models (79) — missing scores show as —
| Benchmark | Category | Sage-S2.1-Flash | Maxwell-V1.0-Flash |
|---|---|---|---|
| OSWorld-G 1/2 models | general_agent | 84.8 | — |
| HMMT Feb 25 1/2 models | stem_reasoning | — | 100 |
| GPQA Diamond 1/2 models | stem_reasoning | — | 86.52 |
| MGSM 1/2 models | multilingual | — | 93 |
| LiveCodeBench v6 (pass@8) 1/2 models | stem_reasoning | — | 93.74 |
| HLE (text-only) 1/2 models | stem_reasoning | — | 24.79 |
| LiveCodeBench v6 1/2 models | stem_reasoning | — | 85.66 |
| Automation-Bench 1/2 models | general_agent | 56.7 | — |
| SWE-bench Pro 1/2 models | coding_agent | 60.6 | — |
| PopQA-300 1/2 models | general_agent | — | 70.1 |
| BFCL-V4 1/2 models | general_agent | 75.5 | — |
| AIME 2025 (mean@8) 1/2 models | stem_reasoning | — | 96.25 |
| HMMT Feb 25 (mean@8) 1/2 models | stem_reasoning | — | 96.25 |
| AA-LCR 1/2 models | long_context | — | 75.6 |
| AIME 26 1/2 models | stem_reasoning | — | 96.67 |
| HMMT Feb 26 (mean@8) 1/2 models | stem_reasoning | — | 91.29 |
| TAU3-Bench 1/2 models | general_agent | — | 67.37 |
| 2WikiMultihopQA-300 1/2 models | general_agent | — | 89.8 |
| GDPVal 1/2 models | general_agent | 78 | — |
| OfficeQA Pro 1/2 models | general_agent | 72.2 | — |
| RoleBench 1/2 models | role_playing | — | 89.92 |
| MLE-bench Lite 1/2 models | coding_agent | 90.3 | — |
| HMMT Nov 25 (mean@8) 1/2 models | stem_reasoning | — | 90.83 |
| MLS-Bench 1/2 models | coding_agent | 42 | — |
| PostTrainBench 1/2 models | coding_agent | 39.2 | — |
| INCLUDE 1/2 models | multilingual | — | 84.7 |
| BabyVision 1/2 models | vision_language | 75.8 | — |
| MMMLU 1/2 models | multilingual | — | 92 |
| OSWorld V2.1 1/2 models | general_agent | 69.3 | — |
| HMMT Nov 25 1/2 models | stem_reasoning | — | 100 |
| BrowseComp-VL 1/2 models | vision_language | 60.5 | — |
| Apex-Agents 1/2 models | general_agent | 61.7 | — |
| Multi-Challenge 1/2 models | instruction_following | — | 72.2 |
| WorkSpaceBench 1/2 models | general_agent | 61.7 | — |
| MMLU-Pro 1/2 models | knowledge | — | 85.46 |
| OSWorld-Verified 1/2 models | general_agent | 85.2 | — |
| HotpotQA-300 1/2 models | general_agent | — | 84.7 |
| Global PIQA 1/2 models | multilingual | — | 92.3 |
| DeepresearchBII 1/2 models | general_agent | 54.2 | — |
| Workspace-Bench-Lite 1/2 models | general_agent | 67.1 | — |
| SpreadsheetBench 2 1/2 models | general_agent | 37.1 | — |
| IFEval 1/2 models | instruction_following | — | 93 |
| HLE (with tools) 1/2 models | stem_reasoning | — | 46.1 |
| MLogiQA 1/2 models | multilingual | — | 81.5 |
| Belebele (5-language subset) 1/2 models | multilingual | — | 92.8 |
| KOR-Bench 1/2 models | reasoning | — | 77.2 |
| WebWalkerQA-300 1/2 models | general_agent | — | 70.5 |
| MMLU-Redux 1/2 models | knowledge | — | 93.59 |
| NL2Repo-Bench 1/2 models | coding_agent | 57.2 | — |
| SkillsBench 1/2 models | general_agent | 62.3 | — |
| WebArena-Verified 1/2 models | general_agent | 67.6 | — |
| CoSER 1/2 models | role_playing | — | 53.54 |
| MCPMark 1/2 models | general_agent | — | 37.3 |
| Harvey Legal Agent 1/2 models | general_agent | 25 | — |
| PaperBench 1/2 models | coding_agent | 88 | — |
| Agents' Last Exam (CLI) 1/2 models | general_agent | 32.4 | — |
| IFBench 1/2 models | instruction_following | — | 64.8 |
| SWE-bench Verified 1/2 models | coding_agent | — | 65.4 |
| LongMemEval 1/2 models | long_context | — | 86.18 |
| IMOAnswerBench 1/2 models | stem_reasoning | — | 95.5 |
| BIG-Bench Hard 1/2 models | reasoning | — | 91.02 |
| AIME 26 (mean@8) 1/2 models | stem_reasoning | — | 95 |
| MuSiQue-300 1/2 models | general_agent | — | 63 |
| AdvancedIF 1/2 models | instruction_following | — | 71.3 |
| MT-RAG 1/2 models | general_agent | — | 65.74 |
| MT-Bench-101 1/2 models | instruction_following | — | 93.3 |
| DeepSWE 1/2 models | coding_agent | 67.9 | — |
| HMMT Feb 26 1/2 models | stem_reasoning | — | 100 |
| AIME 2025 1/2 models | stem_reasoning | — | 100 |
| IMO-AnswerBench (mean@8) 1/2 models | stem_reasoning | — | 86.03 |
| JobBench 1/2 models | general_agent | 58.8 | — |
| RULER (32K context) 1/2 models | long_context | — | 95.81 |
| Bamboogle 1/2 models | general_agent | — | 92.1 |
| C-Eval 1/2 models | knowledge | — | 91 |
| Vision2Web 1/2 models | vision_language | 78 | — |
| TauBench V3 Banking 1/2 models | general_agent | 45.4 | — |
| LongBench v2 1/2 models | long_context | — | 59.4 |
| BigBench Extra Hard 1/2 models | stem_reasoning | — | 52.51 |
| WeaveBench 1/2 models | vision_language | 68.4 | — |
Comparison FAQ
Which model is better, Sage-S2.1-Flash or Maxwell-V1.0-Flash?
Sage-S2.1-Flash wins this comparison with an overall score of 88 vs 49 for Maxwell-V1.0-Flash, leading 4 of 4 shared benchmarks.
Which model is better at General Agents, Sage-S2.1-Flash or Maxwell-V1.0-Flash?
Sage-S2.1-Flash leads the General Agents category (strength score 100/100 across 3 comparable benchmarks).
How reliable is this comparison verdict?
Based on 4 of 83 benchmarks scored by all 2 selected models.