AI Models
Explore the leading LLMs on performance benchmarks, latency, and pricing.
Model
Score (%)
Input Price
Output Price
May/31/2026
MiniMax M3
Overall Score91
Context Window1M
Input Price($/M)$0.30
Output Price($/M)$1.20
Max Output Tokens512k
Max Output Tokens512k
Benchmark Performance
AIMultiple
Holdout
Rank
Category
Benchmarks
Score
Cost Analysis
Blended Price (USD/1M Tokens)
$0.22
From Minimax
Min. Latency
44.71s
From Minimax
FAQ
Consider your primary needs:
Content creation: Focus on AI reasoning and memory scores
Software development: Prioritize AI code performance
Data analysis: Look at text-to-sqlL and AI finance scores
Business automation: Consider Agentic RAG and AI Agents Performance
Factual accuracy: Emphasize low hallucination rates
Content creation: Focus on AI reasoning and memory scores
Software development: Prioritize AI code performance
Data analysis: Look at text-to-sqlL and AI finance scores
Business automation: Consider Agentic RAG and AI Agents Performance
Factual accuracy: Emphasize low hallucination rates
These represent different tiers of OpenAI's GPT-5 family:
GPT-5: Full-featured flagship model
GPT-5 Mini: Optimized for speed and cost while maintaining strong performance
GPT-5 Nano: Ultra-fast, lightweight version for high-volume applications
GPT-5: Full-featured flagship model
GPT-5 Mini: Optimized for speed and cost while maintaining strong performance
GPT-5 Nano: Ultra-fast, lightweight version for high-volume applications
Date suffixes indicate specific training cutoffs or release versions. For example, "claude-3-7-sonnet-20250219" was released on February 19, 2025, helping users track which exact version they're evaluating.
The "32b" in models like "exaone-4.0-32b" refers to 32 billion parameters. Generally, more parameters allow for better performance, but also require more computational resources and cost more to run.
Mini variants: Optimized for speed and cost, typically 65-80% the performance of full models
High variants: Maximum performance configurations, often with increased computational requirements
High variants: Maximum performance configurations, often with increased computational requirements