Model Fortress

Anthropic's most capable generally available model. Step-change improvement in agentic coding over Opus 4.7, with a 1M context window.

Anthropic's most capable generally available model. Step-change improvement in agentic coding over Opus 4.6, with a new tokenizer and 1M context window.

Anthropic's most recent model. Current leader on agentic coding evaluation, Terminal Bench 2.0, and Humanity's Last Exam.

Anthropic's former flagship model combining maximum intelligence with practical performance.

Anthropic's latest Sonnet model with hybrid reasoning, matching near-flagship performance at a fraction of the cost.

Anthropic's latest Sonnet model featuring exceptional performance on coding, analysis, and instruction following.

Anthropic's fastest model with near-frontier intelligence, ideal for high-throughput applications.

Anthropic's previous flagship model with maximum intelligence. Legacy model - consider using Opus 4.5 for new projects.

Anthropic's original Opus 4 model. Legacy model - consider using Opus 4.5 for new projects.

Anthropic's balanced model offering excellent performance across coding, analysis, and complex tasks. A great balance of speed, intelligence, and cost.

OpenAI's GPT-5.5 is a frontier model with a 1M+ context window, reasoning, and broad tool support including web search, file search, image generation, code interpreter, hosted shell, computer use, and MCP. Knowledge cutoff December 1, 2025.

OpenAI's GPT-5.4 is OpenAI's latest frontier model with a 1M+ context window, improved reasoning with xhigh effort support, and enhanced capabilities for coding, agentic tasks, and computer use.

OpenAI's efficient, cost-optimized variant of GPT-5.4 with strong reasoning and multimodal capabilities at a fraction of the cost.

OpenAI's most affordable and fastest variant of GPT-5.4, optimized for high-volume, latency-sensitive applications with minimal cost.

OpenAI's most powerful GPT-5.4 variant with a 1M+ context window, designed for the most demanding reasoning, coding, and agentic tasks with maximum capability.

OpenAI's latest and greatest, GPT 5.2 is OpenAI's flagship model for coding and agentic tasks across industries.

OpenAI's previous flagship model for coding and agentic tasks with configurable reasoning and non-reasoning effort.

OpenAI's former advanced reasoning model with enhanced problem-solving capabilities, deeper understanding, and improved accuracy across complex tasks.

OpenAI's smaller and faster variant of GPT-5, a more cost-efficient alternative. Great for well-defined tasks and precise prompts.

OpenAI's most cost-effective and efficient model in the GPT-5 series, optimized for speed and affordability. Ideal for straightforward tasks and high-volume applications where latency and cost are critical.

OpenAI's reasoning model designed to think before responding. Uses chain-of-thought reasoning for complex tasks in science, coding, and math.

OpenAI's improved version of GPT-4, featuring enhanced reasoning capabilities, better contextual understanding, and increased accuracy across a wide range of tasks.

OpenAI's balanced small model with a massive 1M token context window. Offers great performance at low cost, beating GPT-4o in many benchmarks.

OpenAI's multimodal model optimized for speed and efficiency. Capable of processing text, images, and audio with high accuracy at reduced latency and cost compared to GPT-4 Turbo.

OpenAI's cost-efficient small model. Great for lightweight tasks with fast responses and low cost while maintaining strong capabilities.

OpenAI's GPT-OSS open-weight model designed for powerful reasoning, agentic tasks, and versatile developer use cases.

OpenAI's smaller open-weight model optimized for lower latency and specialized use-cases. Great for edge deployment and faster responses.

OpenAI's GPT-5.3-Codex is the most capable agentic coding model, combining frontier coding performance with reasoning capabilities. Features mid-task steering and 25% faster inference than GPT-5.2-Codex.

OpenAI's GPT-5.2-Codex is an upgraded version of GPT-5.2 optimized for agentic coding tasks in Codex or similar environments.

A version of GPT-5.1-Codex optimized for long-running tasks.

Smaller, more cost-effective, less-capable version of GPT-5.1-Codex.

OpenAI's safety-focused open-weight model with 120B parameters. Designed for content moderation and safe AI applications.

OpenAI's efficient safety-focused open-weight model with 20B parameters. Lightweight content moderation and safety.

Google's lightweight and efficient model in the 3.1 family, optimized for low latency and cost, in preview version.

Google's latest model, in preview version.

Google's latest model with the Flash line's focus on latency, efficiency, and cost, in preview version.

Google's latest Flash-class multimodal model, balancing latency and capability with thinking, function calling, and search grounding.

Google's most capable model for complex reasoning tasks. Features a 1M token context window and strong performance across benchmarks.

Google's fast and cost-effective model with a 1M token context window. Best for high-volume, low-latency tasks and agentic use cases.

Google's most cost-efficient and lowest-latency multimodal model in the 2.5 family, optimized for high-volume classification, simple data extraction, and low-latency tasks.

Google's lightweight open model well-suited for text generation and image understanding. Supports LoRA fine-tuning with 128K context.

Google's lightweight 4B parameter Gemma 3 model. Efficient for simple tasks with vision capabilities.

Google's largest Gemma 3 model with 27B parameters. Strong performance across reasoning, coding, and multilingual tasks with vision support.

Efficient MoE variant of Gemma 4 from Google DeepMind. Multimodal (text + image input), text output.

Dense 30.7B multimodal model from Google DeepMind. Multimodal (text + image input), text output.

xAI's fast coding model trained specifically for agentic coding. Supports reasoning, function calling, structured outputs, and image input. 256K context window.

xAI's Grok 4.3 with 1M token context window. Supports configurable reasoning, function calling, structured outputs, and vision.

xAI's multi-agent model with 2M token context window. Optimized for multi-agent workflows with reasoning capabilities. No client-side or custom tools: Client-side tools (function calling) and custom tools are not currently supported by the multi-agent model variant.

xAI's Grok 4.20 reasoning model with 2M token context window. Supports reasoning, function calling, structured outputs, and vision.

xAI's Grok 4.20 non-reasoning model with 2M token context window. Optimized for speed without reasoning overhead, supports function calling, structured outputs, and vision.

xAI's latest fast reasoning model with massive 2M token context window. Optimized for speed with reasoning capabilities enabled.

xAI's latest fast model with massive 2M token context window. Optimized for speed without reasoning overhead for straightforward tasks.

xAI's Grok 4 model snapshot from July 2025. Premium pricing for maximum capability with 256K context.

xAI's latest and greatest flagship model, offering unparalleled performance in natural language, math, and reasoning - the perfect jack of all trades.

xAI's fast Grok 4 model with reasoning capabilities and massive 2M token context window. Great balance of speed and intelligence.

xAI's fast Grok 4 model without reasoning overhead. Massive 2M context window for processing large documents quickly.

xAI's former flagship reasoning model with strong performance in natural language, math, and coding. Features a 131K context window.

xAI's cost-efficient model with reasoning capabilities. Surprisingly outperforms Grok 3 in many benchmarks while costing 90% less.

DeepSeek's reasoning model trained with reinforcement learning. Excels at math, code, and complex reasoning tasks.

DeepSeek R1 May 2025 update. Reasoning model with strong performance on math, code, and complex reasoning tasks.

DeepSeek's reasoning model distilled from R1 based on Qwen2.5. Outperforms OpenAI o1-mini across various benchmarks with state-of-the-art dense model results.

DeepSeek V3 (March 2024 release) — general-purpose text model with strong reasoning, coding, and instruction-following capabilities.

DeepSeek V3.1 hybrid model on Azure with optional thinking mode, function calling, and a 128K context window.

DeepSeek's latest general-purpose model with improved reasoning, coding, and instruction-following capabilities.

DeepSeek's fast, cost-efficient general-purpose model with 1M context window. Supports tool calling and thinking mode.

DeepSeek's flagship reasoning model with 1M context window. Supports tool calling, extended thinking, and high-quality generation.

Alibaba's Qwen3 model with groundbreaking advancements in reasoning and instruction-following. Excellent for complex tasks and coding.

Alibaba's flagship Qwen3.7 agent model with a 1M-token context window. Hybrid thinking model tuned for coding, agentic workflows, and long-horizon autonomous execution. Supports tool calling and reasoning.

Alibaba's QwQ reasoning model optimized for analytical tasks. Strong performance on math, coding, and logical reasoning benchmarks.

Alibaba's Qwen3 dense 32B model. Strong performance across reasoning, math, and coding tasks.

Alibaba's Qwen3 coding-focused model with 30B total / 3B active parameters using mixture-of-experts architecture.

Alibaba's latest Qwen3 coding model. Cutting-edge code generation and understanding capabilities.

Alibaba's latest Qwen3 model with 80B total / 3B active parameters using mixture-of-experts. Strong general-purpose performance.

Alibaba's Qwen3 vision-language model with 235B total / 22B active parameters. Supports text and image inputs.

Alibaba's Qwen3.6 model with 35B total / 3B active parameters using mixture-of-experts. Strong general-purpose performance.

Alibaba's Qwen3.5 native vision-language model with 122B total / 10B active parameters using a hybrid linear-attention sparse mixture-of-experts architecture.

Alibaba's Qwen3.6 Plus model with a 1M-token context window. Hybrid thinking model supporting text and image inputs, tool calling, structured output, and web search.

Alibaba's Qwen3.6 Flash model with a 1M-token context window. Hybrid thinking model supporting text and image inputs, tool calling, structured output, and web search.

Moonshot AI's trillion-parameter MoE reasoning model (32B activated). Excels at multi-step reasoning with 200+ sequential tool calls. Supports function calling and extended thinking.

Moonshot AI's native multimodal agentic model built on K2. Excels at visual coding, reasoning, and self-directed agent swarm with up to 100 sub-agents.

Moonshot AI's next-gen agentic model built on K2. Long-horizon coding, proactive autonomous execution, and swarm-based task orchestration.

MiniMax's flagship agentic language model with 200K context window. Supports function calling and reasoning.

MiniMax's lightweight MoE model optimized for coding, agentic workflows, and modern application development. 10B activated parameters with strong multilingual code generation.

MiniMax M2.1 optimized for speed at ~100 tokens/second with 200K context window.

MiniMax's latest flagship model with 200K context window, ~60 tokens/second. Supports function calling and thinking.

MiniMax M2.5 optimized for speed at ~100 tokens/second with 200K context window.

MiniMax's latest flagship model with 200K context window and 131K max output. Advanced agentic capabilities with multi-agent collaboration, strong coding, and tool calling.

MiniMax M2.7 optimized for speed with 200K context window and 131K max output.

MiniMax's natively multimodal flagship with a 1M-token context window powered by MiniMax Sparse Attention (MSA). Supports image and PDF input, toggleable thinking, strong agentic tool calling, and long-context reasoning.

Xiaomi's MiMo-V2.5 native omnimodal model with strong agentic capabilities, supporting text, image, video, and audio understanding within a unified architecture. Built on the MiMo-V2-Flash backbone with dedicated vision and audio encoders.

NVIDIA's hybrid MoE model with leading accuracy for multi-agent applications and specialized agentic AI systems.

NVIDIA's compact multimodal Nemotron 3 Nano with 30B total / 3B active MoE parameters and built-in reasoning. Strong at agentic and analytical tasks.

StepFun AI's sparse Mixture-of-Experts model (196B total, 11B active) with a 256K context window, controllable reasoning effort, and native function calling.

ZhipuAI's most capable reasoning model with 128K context window. Supports tool calling, web search, and thinking.

Z-AI's next-generation flagship model for agentic engineering. Stronger coding capabilities and state-of-the-art performance on SWE-Bench Pro.

ZhipuAI's advanced reasoning model with 128K context window. Supports tool calling, web search, and thinking.

ZhipuAI's reasoning model with 128K context window. Supports tool calling, web search, and thinking.

ZhipuAI's efficient reasoning model with 128K context window. Supports tool calling, web search, and thinking.

ZhipuAI's vision-language model with 128K context window. Supports image inputs and thinking.

ZhipuAI's efficient vision-language model with 128K context window. Supports image inputs and thinking.

ZhipuAI's fast and efficient model variant. Optimized for speed while maintaining strong language capabilities.

Meta's latest Llama 4 model with 17B active parameters using mixture-of-experts architecture.

Meta's powerful 70B parameter model quantized to FP8 for fast inference. Excellent for complex reasoning and multilingual tasks.

Meta's compact 3B parameter model optimized for edge deployment and multilingual dialogue. Great balance of speed and capability.

Meta's compact and efficient 1 billion parameter model designed for on-device and edge deployment. Optimized for instruction following with strong performance despite its small size, ideal for resource-constrained environments.

Meta's efficient 8B parameter model optimized for multilingual dialogue. Fast inference with great performance for everyday tasks.

Meta's Llama 4 Maverick with 17B active parameters using mixture-of-experts. Excels at coding, reasoning, and multilingual tasks.

Meta's 11B multimodal model supporting text and image inputs. Optimized for visual reasoning and image understanding tasks.

Meta's largest multimodal model with 90B parameters. Supports text and image inputs with strong reasoning capabilities.

Meta's 70B parameter model with 128K context. Strong performance on reasoning, coding, and multilingual tasks.

Meta's original Llama 3 70B model optimized for dialogue. Strong general-purpose performance across a wide range of tasks.

Meta's efficient Llama 3 8B model optimized for dialogue. Fast inference suitable for lightweight tasks.

Mistral's specialized model for code generation. Optimized for coding tasks including code completion, generation, and explanation.

Mistral's frontier code agents model for solving software engineering tasks. Excels at using tools to explore codebases, editing multiple files, and powering software engineering agents. 256K context.

Mistral's frontier-class multimodal reasoning model.

Mistral's reasoning-focused small model with vision capabilities. Optimized for step-by-step reasoning tasks.

Mistral's flagship 675B parameter model with state-of-the-art reasoning, coding, and multilingual capabilities with vision support.

Mistral's frontier-class multimodal model released May 2025.

Multimodal model from Mistral, released August 2025. Improved tone and performance.

Mistral's best multilingual open source model released July 2024.

Mistral's efficient 24B model optimized for simple tasks with low latency and 128K context. Great for classification, customer support, text generation, and multimodal tasks.

Mistral's efficient 24B model optimized for simple tasks with low latency and 128K context. Great for classification, customer support, text generation, and multimodal tasks.

Mistral's ultra-efficient 3B parameter model with vision support. Designed for edge and low-latency applications.

Mistral's efficient 8B parameter model with vision support. Good balance of capability and speed for moderate tasks.

Mistral's mid-range 14B parameter model with vision support. Enhanced reasoning over smaller Ministrals.

Writer's enterprise-grade model optimized for business content generation, analysis, and transformation.

Writer's most capable enterprise model with enhanced reasoning, analysis, and content generation capabilities.

AI21's Jamba 1.5 Large model with hybrid SSM-Transformer architecture. Excels at long-context understanding and generation.

AI21's efficient Jamba 1.5 Mini model with hybrid SSM-Transformer architecture. Fast and cost-effective for everyday tasks.

Command A is Cohere's most performant model to date, excelling at tool use, agents, retrieval augmented generation (RAG), and multilingual use cases.

Cohere's first multimodal model capable of understanding and interpreting visual data alongside text.

IBM's ultra-efficient micro model. Small but mighty - perfect for simple tasks requiring minimal latency and cost.

Amazon's multimodal model for processing images, video, and text. Can analyze multiple images with 300K context.

Amazon's fastest text-only model optimized for speed and cost. Ideal for text-based tasks requiring low latency with 128K context window.

Amazon's highly capable multimodal model balancing accuracy, speed, and cost. Processes text and images with 300K context.

Amazon's most capable multimodal model for complex reasoning tasks. Processes text and images with up to 1M context window.

Amazon's second-generation lightweight multimodal model. Processes text and images with improved performance over Nova Lite.