BASELINE (2025 leader): The current leader as of Aug 2025 appears to be GPT-4.5 with ~90.2% on MMLU, with Claude 4 and Gemini 2.5 Pro at ~85-86%

Top 7 LLMs Ranked in 2025: GPT-4o, Gemini, Claude & More.

To establish the 2025 baseline:

• On Dec 31, 2025, identify the LLM with the highest average score across the "Core Benchmark Suite" (defined below)

• This becomes the baseline for calculating 10% improvement
  

CORE BENCHMARK SUITE (to avoid cherry-picking):

1. MMLU (general knowledge)

2. HumanEval (coding)

3. GSM8K (math reasoning)

4. ARC-Challenge (scientific reasoning)

5. GPQA (expert-level knowledge)
   

RESOLUTION CRITERIA:

• On Dec 31, 2026, identify the highest-scoring LLM on the same benchmark suite

• Calculate the percentage improvement: (2026_score - 2025_score) / 2025_score × 100

• BET RESOLVES YES if improvement is less than 10%

• BET RESOLVES NO if improvement is 10% or greater
  

KEY DEFINITIONS:

• "LLM": Text-based language models (excludes multimodal-only systems)

• "Publicly available": Model must be accessible via API, open-source, or major consumer platform

• "Score sources": Use official leaderboards (HuggingFace, Papers with Code) or company-reported figures

• "Average": Simple arithmetic mean across the 5 benchmarks
  

EDGE CASES:

• If benchmarks become saturated (>98% scores), substitute with the most widely-adopted replacement benchmark

• If a benchmark is discontinued, use the closest equivalent as determined by academic consensus

• Minimum 3 valid benchmark scores required for inclusion
  

Example calculation:

• 2025 leader: 85% average

• 2026 leader: 92% average

• Improvement: (92-85)/85 = 8.2% → YES (less than 10%)