Category Score V3 leaderboard
LMSpeed Best Multimodal Models
Compare multimodal AI models across perception, OCR, document and spatial understanding, visual reasoning, video, and grounded-action benchmarks in one leaderboard.
Methodology 3.0Methodology
Current answer
Among the currently visible formally ranked models, Qwen3.8 Max has the highest position at global rank 1. Its Category Score is 64.7, with an 80% uncertainty range of ±6.3. 10 visible models have a formal rank. This result applies only to this score run.
Available leaderboard data
- Models shown
- 47
- Formally ranked models
- 10
- Benchmark columns
- 10
- Dimensions with evidence
- 4/4
How to read the benchmark bars
Each bar compares models only within the same benchmark column. Bar lengths are relative to the models shown here; they are not Category Scores and cannot be compared across benchmark columns.
| Rank | Model | LMSpeed score | Perception & OCR | Document & spatial understanding | Visual reasoning | Video & grounded action | Status | Evidence | Updated | ||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| V*undefined models | SimpleVQAundefined models | CharXivundefined models | MMMU-Proundefined models | MathVisionundefined models | ERQAundefined models | MedXpertQA (MM)undefined models | BenchLM Multimodal Grounded scoreundefined models | ScreenSpot Proundefined models | VideoMMMUundefined models | ||||||
| Formally ranked models10 | |||||||||||||||
| 1 | Qwen3.8 MaxQwen | 64.7±6.3 | — | 75.0 | 93.5 | 82.3 | 95.2 | 77.8 | 80.4 | — | 84.5 | 88.7 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 2 | Gemini 3.1 ProGoogle | 55.4±6.6 | — | 72.4 | 80.2 | 83.9 | — | 69.4 | 81.3 | — | 84.4 | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 3 | Qwen3.7 PlusQwen | 55.1±6.3 | — | 81.7 | 85.9 | 79.0 | 90.3 | 69.8 | 71.0 | — | 79.0 | 85.4 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 4 | GPT-5.4OpenAI | 52.9±6.6 | — | 61.1 | 82.8 | 81.2 | — | 65.4 | 77.1 | — | 85.4 | — | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 5 | Qwen3.6 PlusQwen | 50.7±6.5 | 96.9 | — | 81.5 | 78.8 | 88.0 | — | — | — | 68.2 | 84.0 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 6 | Qwen3.5 | 50.2±6.5 | 95.8 | — | 80.8 | 79.0 | 88.6 | — | — | — | 65.6 | 84.7 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 7 | Gemini 3 ProGoogle | 49.9±6.5 | 88.0 | — | 81.4 | 81.0 | 86.6 | — | — | — | 72.7 | 87.6 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 8 | Qwen3.6 27BQwen | 46.3±6.5 | 94.7 | 56.1 | 78.4 | 75.8 | — | 62.5 | — | — | — | 84.4 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 9 | Qwen3.6 35B A3BQwen | 44.1±7.0 | — | 58.9 | 78.0 | 75.3 | — | — | — | — | — | 83.7 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| 10 | Claude Opus 4.5Anthropic | 32.4±6.5 | 67.0 | — | 68.5 | 70.6 | 74.3 | — | — | — | 45.7 | 84.4 | Rated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| Estimated models — unranked18 | |||||||||||||||
| — | Kimi K3MoonshotAI | 65±11.3 | — | — | 91.3 | 81.6 | 94.3 | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.8Anthropic | 59.6±12.0 | — | — | 89.9 | — | — | — | — | — | 87.9 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.8 27BQwen | 57.7±11.3 | — | — | 90.2 | — | 90.0 | 65.5 | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.5 FlashGoogle | 56.3±11.9 | — | — | 84.2 | 83.6 | — | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2.6MoonshotAI | 53.5±9.0 | 96.9 | — | 80.4 | 79.4 | 87.4 | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Kimi K2.5MoonshotAI | 52.7±12.0 | — | — | — | 78.5 | — | — | — | — | — | 86.6 | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiMo-V2.5Xiaomi | 49.1±11.9 | — | — | 81.0 | 77.9 | — | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.5-27BQwen | 49±12.1 | 93.7 | — | — | — | 86.0 | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.6Anthropic | 48.4±11.1 | — | — | — | 77.3 | — | 51.6 | 64.8 | — | 83.1 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.5-122B-A10BQwen | 47.7±9.4 | 93.2 | — | 77.2 | — | 86.2 | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | MiniMax M3MiniMax | 46.9±12.0 | — | — | — | 78.1 | — | — | — | — | — | 84.6 | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Muse Glimmer 30BMeta | 46.4±9.4 | — | — | 78.8 | 74.0 | — | — | — | — | 75.4 | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Inkling SmallThinking Machines | 45.9±11.9 | — | — | 81.3 | 74.0 | — | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Qwen3.5-35B-A3BQwen | 45.8±12.1 | 92.7 | — | — | — | 83.9 | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | InklingThinking Machines | 45.8±11.9 | — | — | 82.0 | 73.5 | — | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.2OpenAI | 42.3±9.0 | 75.9 | — | 82.1 | 79.5 | 83.0 | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Grok 4.20SpaceXAI | 40.2±8.9 | — | 57.4 | 60.9 | 75.2 | — | 54.1 | 65.8 | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Command ACohere | 29.1±11.9 | — | — | 52.7 | 63.0 | — | — | — | — | — | — | Estimated | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| Provisional models — unranked19 | |||||||||||||||
| — | GPT-5.4 ProOpenAI | 73±16.1 | — | — | — | 94.0 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 4.7 MaxAnthropic | 60.7±16.1 | — | — | 91.0 | — | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.6 SolOpenAI | 59.5±16.1 | — | — | — | 83.0 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Step 3.7 FlashStepFun | 59±14.5 | 95.3 | 79.2 | — | — | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.7 FlashGoogle | 57.2±16.1 | — | — | 88.7 | — | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Muse Spark 1.1Meta | 56.8±16.1 | — | — | 88.4 | — | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Sonnet 5Anthropic | 56.6±16.1 | — | — | 88.3 | — | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Opus 5Anthropic | 56.4±17.3 | — | — | — | — | — | — | — | 85.9 | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.5OpenAI | 55.7±16.1 | — | — | — | 81.2 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.6 TerraOpenAI | 54.7±16.1 | — | — | — | 80.7 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.6 FlashGoogle | 51.1±17.3 | — | — | — | — | — | — | — | 72.7 | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.6 LunaOpenAI | 50.3±16.1 | — | — | — | 78.4 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Grok 4.3SpaceXAI | 49.8±16.1 | — | — | — | 78.1 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.4 MiniOpenAI | 47.1±16.1 | — | — | — | 76.6 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.5 Flash-LiteGoogle | 46.9±17.3 | — | — | — | — | — | — | — | 62.3 | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Fable 5Anthropic | 46.3±17.3 | — | — | — | — | — | — | — | 60.8 | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Claude Sonnet 4.6Anthropic | 45.7±16.1 | — | — | 77.4 | — | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | Gemini 3.1 Flash LiteGoogle | 42.6±16.1 | — | — | 73.2 | — | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
| — | GPT-5.4 NanoOpenAI | 31.2±16.1 | — | — | — | 66.1 | — | — | — | — | — | — | Provisional | undefined/4 dimensions · undefined families | Aug 28, 2026 |
What this leaderboard measures
Which AI model is better suited to image and document tasks?
This leaderboard covers perception, OCR, document and spatial understanding, visual reasoning, video, and action. Supported input types can differ between models.
Four capability dimensions
The four dimensions come from the category blueprint. Available data may cover only some of them. A dimension without evidence is not presented as a verified capability.
Perception & OCR
2 benchmark columns currently provide evidence for this dimension.
- V*
- SimpleVQA
Document & spatial understanding
1 benchmark columns currently provide evidence for this dimension.
- CharXiv
Visual reasoning
5 benchmark columns currently provide evidence for this dimension.
- MMMU-Pro
- MathVision
- ERQA
- MedXpertQA (MM)
- BenchLM Multimodal Grounded score
Video & grounded action
2 benchmark columns currently provide evidence for this dimension.
- ScreenSpot Pro
- VideoMMMU
Multimodal tasks this page can help with
- Document extraction that reads text, tables, layout, and page relationships.
- Image questions that combine visual details with written reasoning.
- Video or visual agents that understand a process and use visual information to act.
How to choose a model with this leaderboard
- Step 1
Check the rating status first
Only Rated models receive a rank. Estimated and Provisional models do not have a formal position.
- Step 2
Review uncertainty and evidence
When scores are close, do not rely on rank alone. Check uncertainty, dimension coverage, and benchmark count.
- Step 3
Test the real task last
A leaderboard cannot replace your own test. Check quality, speed, price, context, and provider limits together.
Confirm supported files and input types first. Also check OCR languages, file limits, speed, cost, and privacy.
Rating status guide
Rated
Rated means the evidence and overlap rules are met. The model can receive a formal rank.
Estimated
Estimated means there is useful evidence, but it is not enough for a formal rank.
Provisional
Provisional means evidence is limited or dimension and benchmark-family coverage is below the estimated threshold. Use the result only as an early signal.
Benchmarks and evidence sources
Evidence source names and benchmark groups come from the currently available score data. One source may contribute several benchmarks.
BenchLM
BenchLM Multimodal Grounded score, CharXiv, ERQA, MathVision, MedXpertQA (MM), MMMU-Pro, ScreenSpot Pro, SimpleVQA, V*, and VideoMMMU
How the multimodal model ranking is built
LMSpeed combines eligible third-party benchmarks inside four fixed capability dimensions. Rated models meet the evidence and overlap requirements for a formal rank; Estimated and Provisional models remain visible without receiving a rank.
Read the Category Score methodologyLeaderboard limits
Category Scores use the third-party benchmarks currently included by LMSpeed. Tests can use different data, prompts, and scoring rules. The result is not permanent and cannot represent every real task. Test important choices with your own data and workflow.
Frequently asked questions
Which visible model has the highest formal rank now?
Among the currently visible models, Qwen3.8 Max has the highest formal position at global rank 1. Its Category Score is 64.7. 10 visible models meet the formal ranking rules. This result applies only to the run date and methodology version shown on the page.
Can I compare scores across different categories?
No. Each category uses different capability dimensions and evidence. A Category Score is comparable only inside the same leaderboard. Review the matching category for each task.
Are Estimated and Provisional models still useful?
They can help you find candidates, but their evidence is not complete enough for a formal rank. Review coverage and uncertainty, then test the model on a real task.
How often does the leaderboard update?
The leaderboard updates after a new completed score run is published. The current run date and methodology version appear above. LMSpeed does not promise a fixed daily or weekly schedule.
Is the number one model always best for me?
No. Your result also depends on speed, price, context length, tool support, region, and provider limits. Use the leaderboard to narrow the field, then run your own test.
Does a high multimodal score support every image and video format?
No. Benchmark scores and product feature support are different. Confirm file formats, size limits, video support, OCR languages, and provider APIs before choosing a model.