Ornith-1.0-9B vs Gemma4-12B vs LFM3-Omni-2B
New ComparisonScores are fetched live from the current database on every visit — nothing is cached.
Winner
Ornith-1.0-9B
Ornith-1.0-9B scores higher overall across category strength profiles (76 vs 50 for LFM3-Omni-2B); only 0 benchmarks are directly comparable.
Ornith-1.0-9B
75.5
LFM3-Omni-2B
49.9
Gemma4-12B
43.1
Only 0 of 70 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.
Benchmark Scores Comparison
Ornith-1.0-9B
Gemma4-12B
LFM3-Omni-2B
Ornith-1.0-9B
9.0B params
Score: 100.0
Gemma4-12B
11.9B params
Score: 100.0
LFM3-Omni-2B
2.2B params
Coverage matrix
3 models × 70 benchmarks — hover a benchmark name for its category
Shading = score relative to the best score among the selected models:
Low
Mid
High
Missing (—)
The totals row shows how many of the 70 benchmarks each model has scores for.
| Benchmark | Gemma4-12B Google | Ornith-1.0-9B DeepReinforce AI | LFM3-Omni-2B LiquidAI |
|---|---|---|---|
| RealWorldQA (spoken) | — | — | 64.68 |
| GPQA Diamond | 78.8 | — | — |
| MathVista (mini) | — | — | 62.5 |
| MMBench EN-DEV v1.1 | — | — | 73.68 |
| LiveCodeBench v6 | 72 | — | — |
| SWE-bench Pro | 21.9 | 42.9 | — |
| BFCL-V4 | 37.4 | — | — |
| RealWorldQA | — | — | 66.54 |
| EmbSpatialBench | — | — | 70.55 |
| AA-LCR | 55.3 | — | — |
| MathVision | 79.7 | — | — |
| AIME 26 | 82.1 | — | — |
| FLEURS | 0.069 | — | 6.72 |
| VITA-Bench | 12.4 | — | — |
| TAU3-Bench | 13.3 | — | — |
| MMStar | — | — | 55.67 |
| WorldSense | — | — | 47.1 |
| SWE Atlas - TW | — | 15.3 | — |
| EmbSpatial (spoken) | — | — | 67.59 |
| SWE Atlas - QnA | — | 17.9 | — |
| CoVoST | 38.5 | — | — |
| NL2Repo | 10.3 | 27.2 | — |
| MMMU-Pro | 69.1 | — | 27.28 |
| MMMLU | 83.4 | — | — |
| Claw-Eval Avg | 32.5 | 63.1 | — |
| Daily-Omni | — | — | 70.09 |
| InfoVQA | — | — | 64.08 |
| MMLU-Pro | 77.2 | — | — |
| CodeForces | 1659 | — | — |
| IFEval | 94.8 | — | — |
| OmniBench (overall) | — | — | 46.32 |
| POPE | — | — | 89.34 |
| MRCR v2 128K (8-needle) | 43.4 | — | — |
| BrowseComp | 10 | — | — |
| MedXpertQA-MM | 48.7 | — | — |
| English TTS UTMOSv2 | — | — | 3.8 |
| SWE-bench Multilingual | 32.5 | 52 | — |
| English ASR WER | — | — | 6.36 |
| RoboSpatial | — | — | 49.14 |
| AV-SpeakerBench | — | — | 46.79 |
| MMAR | — | — | 51.4 |
| IFBench | 73.5 | — | — |
| SWE-bench Verified | 44.2 | 69.4 | — |
| IMOAnswerBench | 57.2 | — | — |
| Humanity's Last Exam | 13.1 | — | — |
| ERQA | — | — | 38.5 |
| VideoMME (w/o sub.) | — | — | 57.7 |
| ERQA (spoken) | — | — | 39.43 |
| TAU2-Bench | 69 | — | — |
| OmniBench (speech) | — | — | 48.25 |
| Terminal-Bench 2.1 (Terminus-2) | 21 | 43.1 | — |
| MMStar (spoken) | — | — | 51.79 |
| OmniDocBench 1.5 | 0.164 | — | — |
| OCRBench | — | — | 806 |
| HMMT Feb 26 | 65.6 | — | — |
| GPQA | 72.8 | — | — |
| POPE (spoken) | — | — | 83.78 |
| RefCOCO (avg) | — | — | 73.55 |
| VoiceBench | — | — | 64.73 |
| SEEDBench (image) | — | — | 75.07 |
| MCP-Atlas | 30.5 | — | — |
| Tool calling (internal eval) | — | — | 52.2 |
| DocVQA | — | — | 92.47 |
| SWE Atlas - RF | — | 16.6 | — |
| Terminal-Bench 2.1 (Claude Code) | — | 40.6 | — |
| BigBench Extra Hard | 53 | — | — |
| Gaokao 2026 | 130.6 | — | — |
| SciCode | 39.8 | — | — |
| MMMU | — | — | 42.44 |
| MVBench | — | — | 58.03 |
| Covered | 35/70 | 10/70 | 33/70 |
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 |
|---|---|---|---|---|---|
| Ornith-1.0-9B Winner | 75.5 | — | 10 of 70 (14.0%) | Agentic Coding | Agentic Coding |
| LFM3-Omni-2B | 49.9 | — | 33 of 70 (47.0%) | Multilingual | Vision & Language |
| Gemma4-12B | 43.1 | — | 35 of 70 (50.0%) | Vision & Language | Multilingual |
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 | Ornith-1.0-9B | LFM3-Omni-2B | Gemma4-12B |
|---|---|---|---|
| Vision & Language 15 benchmarks | — | 2.1 | 100 |
| Agentic Coding 10 benchmarks | 100 | — | 37.1 |
| Multilingual 3 benchmarks | — | 100 | 0 |
Partial coverage benchmarks
Scored by at least one but not all selected models (70) — missing scores show as —
| Benchmark | Category | Ornith-1.0-9B | Gemma4-12B | LFM3-Omni-2B |
|---|---|---|---|---|
| RealWorldQA (spoken) 1/3 models | vision_language | — | — | 64.68 |
| GPQA Diamond 1/3 models | stem_reasoning | — | 78.8 | — |
| MathVista (mini) 1/3 models | vision_language | — | — | 62.5 |
| MMBench EN-DEV v1.1 1/3 models | vision_language | — | — | 73.68 |
| LiveCodeBench v6 1/3 models | stem_reasoning | — | 72 | — |
| SWE-bench Pro 2/3 models | coding_agent | 42.9 | 21.9 | — |
| BFCL-V4 1/3 models | general_agent | — | 37.4 | — |
| RealWorldQA 1/3 models | vision_language | — | — | 66.54 |
| EmbSpatialBench 1/3 models | spatial_intelligence | — | — | 70.55 |
| AA-LCR 1/3 models | long_context | — | 55.3 | — |
| MathVision 1/3 models | vision_language | — | 79.7 | — |
| AIME 26 1/3 models | stem_reasoning | — | 82.1 | — |
| FLEURS 2/3 models | multilingual | — | 0.069 | 6.72 |
| VITA-Bench 1/3 models | general_agent | — | 12.4 | — |
| TAU3-Bench 1/3 models | general_agent | — | 13.3 | — |
| MMStar 1/3 models | vision_language | — | — | 55.67 |
| WorldSense 1/3 models | multimodal | — | — | 47.1 |
| SWE Atlas - TW 1/3 models | coding_agent | 15.3 | — | — |
| EmbSpatial (spoken) 1/3 models | vision_language | — | — | 67.59 |
| SWE Atlas - QnA 1/3 models | coding_agent | 17.9 | — | — |
| CoVoST 1/3 models | multilingual | — | 38.5 | — |
| NL2Repo 2/3 models | coding_agent | 27.2 | 10.3 | — |
| MMMU-Pro 2/3 models | vision_language | — | 69.1 | 27.28 |
| MMMLU 1/3 models | multilingual | — | 83.4 | — |
| Claw-Eval Avg 2/3 models | coding_agent | 63.1 | 32.5 | — |
| Daily-Omni 1/3 models | multimodal | — | — | 70.09 |
| InfoVQA 1/3 models | multimodal | — | — | 64.08 |
| MMLU-Pro 1/3 models | knowledge | — | 77.2 | — |
| CodeForces 1/3 models | stem_reasoning | — | 1659 | — |
| IFEval 1/3 models | instruction_following | — | 94.8 | — |
| OmniBench (overall) 1/3 models | multimodal | — | — | 46.32 |
| POPE 1/3 models | vision_language | — | — | 89.34 |
| MRCR v2 128K (8-needle) 1/3 models | long_context | — | 43.4 | — |
| BrowseComp 1/3 models | general_agent | — | 10 | — |
| MedXpertQA-MM 1/3 models | vision_language | — | 48.7 | — |
| English TTS UTMOSv2 1/3 models | multimodal | — | — | 3.8 |
| SWE-bench Multilingual 2/3 models | coding_agent | 52 | 32.5 | — |
| English ASR WER 1/3 models | multimodal | — | — | 6.36 |
| RoboSpatial 1/3 models | spatial_intelligence | — | — | 49.14 |
| AV-SpeakerBench 1/3 models | multimodal | — | — | 46.79 |
| MMAR 1/3 models | multimodal | — | — | 51.4 |
| IFBench 1/3 models | instruction_following | — | 73.5 | — |
| SWE-bench Verified 2/3 models | coding_agent | 69.4 | 44.2 | — |
| IMOAnswerBench 1/3 models | stem_reasoning | — | 57.2 | — |
| Humanity's Last Exam 1/3 models | stem_reasoning | — | 13.1 | — |
| ERQA 1/3 models | spatial_intelligence | — | — | 38.5 |
| VideoMME (w/o sub.) 1/3 models | video_understanding | — | — | 57.7 |
| ERQA (spoken) 1/3 models | vision_language | — | — | 39.43 |
| TAU2-Bench 1/3 models | general_agent | — | 69 | — |
| OmniBench (speech) 1/3 models | multimodal | — | — | 48.25 |
| Terminal-Bench 2.1 (Terminus-2) 2/3 models | coding_agent | 43.1 | 21 | — |
| MMStar (spoken) 1/3 models | vision_language | — | — | 51.79 |
| OmniDocBench 1.5 1/3 models | document_understanding | — | 0.164 | — |
| OCRBench 1/3 models | document_understanding | — | — | 806 |
| HMMT Feb 26 1/3 models | stem_reasoning | — | 65.6 | — |
| GPQA 1/3 models | reasoning | — | 72.8 | — |
| POPE (spoken) 1/3 models | vision_language | — | — | 83.78 |
| RefCOCO (avg) 1/3 models | spatial_intelligence | — | — | 73.55 |
| VoiceBench 1/3 models | multimodal | — | — | 64.73 |
| SEEDBench (image) 1/3 models | vision_language | — | — | 75.07 |
| MCP-Atlas 1/3 models | general_agent | — | 30.5 | — |
| Tool calling (internal eval) 1/3 models | multimodal | — | — | 52.2 |
| DocVQA 1/3 models | multimodal | — | — | 92.47 |
| SWE Atlas - RF 1/3 models | coding_agent | 16.6 | — | — |
| Terminal-Bench 2.1 (Claude Code) 1/3 models | coding_agent | 40.6 | — | — |
| BigBench Extra Hard 1/3 models | stem_reasoning | — | 53 | — |
| Gaokao 2026 1/3 models | — | — | 130.6 | — |
| SciCode 1/3 models | — | — | 39.8 | — |
| MMMU 1/3 models | vision_language | — | — | 42.44 |
| MVBench 1/3 models | video_understanding | — | — | 58.03 |
Comparison FAQ
Which model scores highest in this comparison?
Ornith-1.0-9B scores higher overall across category strength profiles (76 vs 50 for LFM3-Omni-2B); only 0 benchmarks are directly comparable.
Which model is strongest at Agentic Coding?
Ornith-1.0-9B leads the Agentic Coding category (strength score 100/100 across 6 comparable benchmarks).
How reliable is this comparison verdict?
Only 0 of 70 benchmarks are scored by every selected model — the verdict leans on category strength profiles and is indicative.