Latest Models by Company
Explore the most recent releases from major AI companies
Anthropic
Safety-first AI • Constitutional AI approach
Claude 4.1 Opus
Anthropic • August 2025
Training Data
Up to mid 2025
Claude 4.1 Opus
August 2025
Parameters
~500 billion
Training Method
Advanced Constitutional AI
Context Window
200,000 tokens
Knowledge Cutoff
July 2025
Key Features
Latest Improvements • Enhanced Performance • Advanced Reasoning
Capabilities
Reasoning: Maximum
Coding: Outstanding
Performance: Peak
What's New in This Version
Latest iteration with enhanced performance across all benchmarks
Anthropic's most advanced model with enhanced capabilities and latest improvements
OpenAI
Innovation leader • GPT family
GPT-5
OpenAI • August 2025
Training Data
Up to September 2024
GPT-5
August 2025
Parameters
~1.7 trillion
Training Method
Unified Architecture with Reasoning
Context Window
256,000 tokens
Knowledge Cutoff
September 2024
Key Features
Unified Intelligence • Advanced Reasoning • Native Multimodal • Real-time Router
Capabilities
Reasoning: Outstanding
Multimodal: Outstanding
Accuracy: Exceptional
What's New in This Version
80% fewer factual errors than o3, unified reasoning and multimodal in single system
OpenAI's unified intelligence model combining advanced reasoning with multimodal capabilities
Gemini AI • Search integration
Gemini 2.5 Pro
Google • March 2025
Training Data
Up to late 2024
Gemini 2.5 Pro
March 2025
Parameters
~800 billion
Training Method
Thinking-Enhanced Transformer
Context Window
2,000,000 tokens
Knowledge Cutoff
February 2025
Key Features
Deep Thinking • Advanced Math • Agentic Code
Capabilities
Math/Science: State-of-the-art
Coding: 63.8% SWE-Bench
Reasoning: Industry-Leading
What's New in This Version
18.8% on Humanity's Last Exam, state-of-the-art on GPQA and AIME 2025
Google's most intelligent AI model with advanced thinking capabilities
Meta
Llama models • Open source focus
Llama 4 Behemoth
Meta • April 2025
Training Data
Up to August 2024
Llama 4 Behemoth
April 2025
Parameters
~2 trillion (288B active)
Training Method
Mixture of Experts
Context Window
1,000,000 tokens
Knowledge Cutoff
August 2024
Key Features
Open Source • Massive MoE Architecture • Multimodal
Capabilities
Reasoning: Outstanding
STEM: Outstanding
Complex Tasks: Outstanding
What's New in This Version
Massive MoE model with 16 experts designed for complex reasoning tasks
Meta's flagship multimodal model with massive MoE architecture (288B active parameters)
Mistral AI
Mixture of experts • French precision
Magistral Medium
Mistral AI • June 2025
Training Data
Up to June 2025
Magistral Medium
June 2025
Parameters
~200 billion
Training Method
Reasoning-focused Training
Context Window
40,000 tokens
Knowledge Cutoff
June 2025
Key Features
Advanced Reasoning • Multi-step Logic • Multilingual Excellence
Capabilities
Reasoning: Outstanding (73.6% AIME2024)
Logic: Outstanding
Multilingual: Excellent
What's New in This Version
First reasoning model from Mistral with traceable thought processes
Mistral's flagship reasoning model with advanced multi-step logic capabilities
DeepSeek
Chinese innovation • MoE architecture
DeepSeek-R1-0528
DeepSeek • May 2025
Training Data
Up to May 2025
DeepSeek-R1-0528
May 2025
Parameters
671 billion (37B active)
Training Method
Enhanced Reinforcement Learning
Context Window
128,000 tokens
Knowledge Cutoff
May 2025
Key Features
Enhanced Reasoning • System Prompts • JSON Output • Function Calling • Reduced Hallucinations
Capabilities
Reasoning: Outstanding (87.5% AIME)
Math: Outstanding
Function Calling: Outstanding
What's New in This Version
Major upgrade: 87.5% AIME accuracy (vs 70%), deeper thinking (23K vs 12K tokens), 45-50% fewer hallucinations
DeepSeek's upgraded reasoning model with 87.5% AIME accuracy and significantly reduced hallucinations
Moonshot AI
Long context expert • Kimi AI chatbot
Kimi K2
Moonshot AI • July 2025
Training Data
Up to July 2025
Kimi K2
July 2025
Parameters
1 trillion (32B active)
Training Method
Mixture of Experts with MuonClip optimizer
Context Window
128,000 tokens
Knowledge Cutoff
July 2025
Key Features
Open Source (MIT) • Agentic Intelligence • 384 Experts MoE • Native Tool Use
Capabilities
Coding: Outstanding (71.6% SWE-bench)
Agentic Tasks: Outstanding (65.8%)
LiveCodeBench: 53.7%
What's New in This Version
Purpose-built for agentic workflows with native MCP support and multi-step tool interactions
Moonshot AI's trillion-parameter agentic model with superior coding and reasoning
xAI
Real-time knowledge • Grok personality
Grok-4
xAI • July 2025
Training Data
Up to July 2025
Grok-4
July 2025
Parameters
~2.4 trillion (estimated)
Training Method
Multi-agent Advanced Reinforcement Learning
Context Window
1,000,000 tokens
Knowledge Cutoff
Real-time via X
Key Features
Multi-agent Architecture • Native Tool Use • Real-time Search • 200K GPU Cluster
Capabilities
Reasoning: Outstanding (25.4% HLE)
Tool Use: Outstanding
Real-time: Outstanding
What's New in This Version
Multi-agent reasoning with 44.4% on Humanity's Last Exam with tools, PhD-level performance
xAI's most advanced model with multi-agent architecture and breakthrough reasoning
How to Use LLM Bento
Explore Models
Browse the latest LLM models by company using our elegant glass morphism interface
Compare & Learn
Use the timeline and comparison tools to understand technical differences
Understand Concepts
Learn LLM terminology with clear explanations and industry context