Gemini 3.1 Pro vs GPT-5.5: Analyzing a Complex Medical Infographic
The Prompt
Here's a medical infographic showing global diabetes prevalence rates by region from 2020-2025, with treatment cost breakdowns and projected trends through 2030. Analyze this image: identify every data point, flag any inconsistencies between the chart and the text labels, and produce a structured summary table with regional breakdowns, year-over-year growth rates, and the top 3 actionable takeaways for a healthcare policy researcher.
This prompt tests multimodal analysis at its hardest — extracting precise numerical data from a dense infographic, cross-referencing visual elements with text, performing calculations, and synthesizing actionable insights. It requires vision accuracy, numerical reasoning, and structured output generation simultaneously.
Gemini 3.1 ProGPT-5.5
## Infographic Analysis: Global Diabetes Prevalence 2020-2025\n\n### Data Extraction\n\nI've identified 42 discrete data points across the infographic. Here's the structured breakdown:\n\n| Region | 2020 Prevalence | 2025 Prevalence | YoY Growth | Treatment Cost (2025) |\n|--------|----------------|-----------------|------------|----------------------|\n| North America | 10.7% | 12.1% | +2.6% CAGR | $4,820/patient |\n| Europe | 8.9% | 9.8% | +1.9% CAGR | $3,650/patient |\n| South-East Asia | 8.7% | 11.3% | +5.4% CAGR | $890/patient |\n| Middle East & N. Africa | 12.2% | 14.5% | +3.5% CAGR | $1,420/patient |\n| Sub-Saharan Africa | 3.1% | 4.6% | +8.2% CAGR | $340/patient |\n| Western Pacific | 9.8% | 11.9% | +3.9% CAGR | $2,100/patient |\n\n### Inconsistencies Found\n\n1. **South-East Asia bar chart shows 10.8% for 2025** but the text label reads 11.3% — the text label appears to include updated IDF Atlas data while the chart uses the original dataset\n2. **Treatment cost pie chart for Europe** segments total 103.2% — likely a rounding error in the sub-categories (hospitalization 41%, medication 38%, monitoring 24.2%)\n3. **2030 projection line for Sub-Saharan Africa** intersects the Middle East line at 2028, but the accompanying text states convergence happens in 2029\n\n### 2030 Projections (extracted from trend lines)\n- Sub-Saharan Africa: fastest growth trajectory, projected 7.8% by 2030\n- South-East Asia: projected 14.1%, overtaking MENA\n- Global average: projected 11.2% (up from 9.3% current)\n\n### Top 3 Actionable Takeaways\n\n1. **Sub-Saharan Africa's 8.2% CAGR demands immediate infrastructure investment** — the region has the fastest growth rate but the lowest per-patient treatment spend ($340). The cost-prevalence gap will become unsustainable by 2028.\n\n2. **South-East Asia's data discrepancy (10.8% vs 11.3%) signals the problem is accelerating faster than official projections** — policy should use the higher figure for resource planning.\n\n3. **Europe's 103.2% cost breakdown suggests hidden costs aren't captured** — the monitoring category (24.2%) likely includes digital health tools that weren't in the original methodology. Recommend a cost audit before setting 2027 budget baselines.
GPT-5.5GPT-5.5
## Analysis of Diabetes Prevalence Infographic\n\n### Overview\nThis infographic presents global diabetes data across 6 major regions from 2020 to 2025, with treatment costs and 2030 projections.\n\n### Regional Data Summary\n\n| Region | 2020 | 2025 | Growth | Cost/Patient |\n|--------|------|------|--------|--------------|\n| North America | 10.7% | 12.1% | +1.4pp | $4,820 |\n| Europe | 8.9% | 9.8% | +0.9pp | $3,650 |\n| South-East Asia | 8.7% | 11.3% | +2.6pp | $890 |\n| MENA | 12.2% | 14.5% | +2.3pp | $1,420 |\n| Sub-Saharan Africa | 3.1% | 4.6% | +1.5pp | $340 |\n| Western Pacific | 9.8% | 11.9% | +2.1pp | $2,100 |\n\n### Observations\n- MENA region has the highest current prevalence at 14.5%\n- Sub-Saharan Africa shows rapid percentage growth despite low absolute numbers \n- North America has the highest per-patient costs by a significant margin\n\n### Potential Inconsistency\nThe South-East Asia figure appears to differ slightly between the chart visualization and the text annotation. The text reads 11.3% but the bar appears closer to 10.8%.\n\n### Key Takeaways\n1. **Spending disparity is stark**: North America spends 14x more per patient than Sub-Saharan Africa, yet both regions show significant prevalence growth\n2. **MENA needs targeted intervention** as the highest-prevalence region\n3. **South-East Asia's rapid growth** makes it a priority for preventive programs
🔍 Analysis
Gemini 3.1 Pro dominates this multimodal analysis test. Both models extracted the core data table correctly, but the differences in depth and analytical rigor are significant.\n\nData Extraction: Gemini identified 42 discrete data points and calculated CAGR growth rates, while GPT-5.5 reported simple percentage-point changes. For a policy researcher, CAGR is far more useful for projecting future resource needs.\n\nInconsistency Detection: Gemini found 3 inconsistencies (the South-East Asia data mismatch, Europe's cost pie chart totaling 103.2%, and the projection line intersection error). GPT-5.5 found only the South-East Asia discrepancy. The cost pie chart catch is particularly impressive — it requires adding up segments from a secondary visual element.\n\nProjection Analysis: Gemini extracted specific 2030 projection numbers from the trend lines and identified when regional lines would cross. GPT-5.5 didn't attempt to extract projection data.\n\nActionable Insights: Gemini's takeaways directly tied back to specific data anomalies and had concrete implications (\"use the higher figure,\" \"conduct cost audit before 2027 budgets\"). GPT-5.5's takeaways were accurate but generic — they described the data rather than advising on it.\n\nThis result aligns with benchmarks: Gemini 3.1 Pro scores 82.8 on multimodal tasks vs GPT-5.5's 70.4. Its natively multimodal architecture — processing text, images, and data in a unified pipeline — gives it a structural advantage over GPT-5.5's modular approach for dense visual reasoning tasks. GPT-5.5 remains stronger for pure text reasoning (85 vs 77.1), but when the image IS the data source, Gemini is the clear pick.
"We gave Gemini 3.1 Pro and GPT-5.5 a dense medical infographic and asked them to extract every data point, find inconsistencies, and produce policy-ready insights. Gemini found 3 errors GPT missed entirely. Full breakdown:"