Multimodal · Benchmark profile
A harder multimodal benchmark for frontier models that combines text with images, diagrams, charts, and academic visual reasoning tasks.
Data verified 27 Jul 2026 · Methodology 1.8.0
Benchmark score on Massive Multi-discipline Multimodal Understanding Pro
GPT-5.4 Pro leads at 94%, followed by Gemini 3.1 Pro (83.9%) and Gemini 3.5 Flash (83.6%).
10 ranked models · higher is better · labels show rank and score
| Model | Rank | Provider | Score |
|---|---|---|---|
| GPT-5.4 Pro | #1 | OpenAI | 94% |
| Gemini 3.1 Pro | #2 | 83.9% | |
| Gemini 3.5 Flash | #3 | 83.6% | |
| GPT-5.6 Sol | #4 | OpenAI | 83% |
| Kimi K3 | #5 | Moonshot AI | 81.6% |
| GPT-5.4 | #6 | OpenAI | 81.2% |
| GPT-5.5 | #7 | OpenAI | 81.2% |
| Gemini 3 Pro | #8 | 81% | |
| GPT-5.6 Terra | #9 | OpenAI | 80.7% |
| Muse Spark | #10 | Meta | 80.4% |
One best score per model · higher is better
| Rank | Model | Provider | License | Evidence use | Score |
|---|---|---|---|---|---|
| #1 | GPT-5.4 Pro gpt-5-4-pro | OpenAI | closed | Estimated reference | 94% |
| #2 | Gemini 3.1 Pro gemini-3-1-pro | closed | Estimated reference | 83.9% | |
| #3 | Gemini 3.5 Flash gemini-3.5-flash | closed | Estimated reference | 83.6% | |
| #4 | GPT-5.6 Sol gpt-5.6-sol | OpenAI | closed | Reference only | 83% |
| #5 | Kimi K3 kimi-k3 | Moonshot AI | closed | Reference only | 81.6% |
| #6 | GPT-5.4 gpt-5.4 | OpenAI | closed | Reference only | 81.2% |
| #7 | GPT-5.5 gpt-5.5 | OpenAI | closed | Estimated reference | 81.2% |
| #8 | Gemini 3 Pro gemini-3-pro | closed | Reference only | 81% | |
| #9 | GPT-5.6 Terra gpt-5.6-terra | OpenAI | closed | Estimated reference | 80.7% |
| #10 | Muse Spark muse-spark | Meta | closed | Reference only | 80.4% |
| #11 | GPT-5.2 gpt-5.2 | OpenAI | closed | Estimated reference | 79.5% |
| #12 | Kimi K2.6 kimi-k2.6 | Moonshot AI | open | Estimated reference | 79.4% |
| #13 | Qwen3.5 397B qwen3.5-397b-a17b | Alibaba Cloud | open | Estimated reference | 79% |
| #14 | Qwen3.7-Plus qwen3.7-plus | Alibaba Cloud | closed | Reference only | 79% |
| #15 | Qwen3.6 Plus qwen3-6-plus | Alibaba Cloud | closed | Reference only | 78.8% |
| #16 | Kimi K2.5 kimi-k2.5 | Moonshot AI | open | Reference only | 78.5% |
| #17 | Kimi K2.5 (Reasoning) kimi-k2-5-reasoning | Moonshot AI | closed | Estimated reference | 78.5% |
| #18 | GPT-5.6 Luna gpt-5.6-luna | OpenAI | closed | Estimated reference | 78.4% |
| #19 | Grok 4.3 grok-4.3 | xAI | closed | Reference only | 78.1% |
| #20 | MiniMax M3 MiniMax-M3 | MiniMax | open | Reference only | 78.1% |
| #21 | MiMo-V2.5 mimo-v2.5 | Xiaomi | open | Estimated reference | 77.9% |
| #22 | Claude Opus 4.6 claude-opus-4-6 | Anthropic | closed | Reference only | 77.3% |
| #23 | Gemma 4 31B gemma-4-31b | open | Reference only | 76.9% | |
| #24 | GPT-5.4 mini gpt-5.4-mini | OpenAI | closed | Estimated reference | 76.6% |
| #25 | Qwen3.6 27B qwen3.6-27b | Alibaba Cloud | open | Estimated reference | 75.8% |
| #26 | Qwen3.6-35B-A3B qwen3-6-35b-a3b | Alibaba Cloud | open | Estimated reference | 75.3% |
| #27 | Grok 4.20 grok-4.20 | xAI | closed | Estimated reference | 75.2% |
| #28 | Gemma 4 26B A4B gemma-4-26b-a4b | open | Reference only | 73.8% | |
| #29 | Inkling inkling | Thinking Machines Lab | open | Reference only | 73.5% |
| #30 | Interfaze Beta interfaze-beta | Interfaze | closed | Estimated reference | 71.1% |
| #31 | Claude Opus 4.5 claude-opus-4-5 | Anthropic | closed | Reference only | 70.6% |
| #32 | Gemma 4 12B gemma-4-12b | open | Estimated reference | 69.1% | |
| #33 | GPT-5.4 nano gpt.5.4.nano | OpenAI | closed | Reference only | 66.1% |
| #34 | Command A+ command-a-plus | Cohere | open | Estimated reference | 63% |
The top of this snapshot is led by GPT-5.4 Pro at 94%; third place is 10.4 points behind. The top-10 spread is 13.6 points.
About Massive Multi-discipline Multimodal Understanding Pro
A harder multimodal benchmark for frontier models that combines text with images, diagrams, charts, and academic visual reasoning tasks. Results stay tied to the exact model variant and evaluation system. Multiple systems for the same model use the best published score on this page; overall Lumina scoring uses the median of ranking-eligible rows.
Open benchmark source ↗FAQ
A harder multimodal benchmark for frontier models that combines text with images, diagrams, charts, and academic visual reasoning tasks.
GPT-5.4 Pro by OpenAI currently leads with 94%.
34 models in the LuminaBench cohort have a qualifying score on this benchmark.
Related
