Explore AI Models

Gemini 3.1 Flash Lite Preview is Google’s efficient high-volume model optimized for speed and cost. It delivers strong performance in translation, data extraction, and code completion while supporting adjustable reasoning levels for flexible workloads.

Frontier reasoning model for advanced agentic systems and complex coding. Delivers stronger engineering performance, better autonomous execution, improved token efficiency, and long-context stability, with a 1M-token context and full multimodal support (text, image, video, audio, code).

High-speed, high-value thinking model designed for agentic workflows and coding. Delivers near-Pro level reasoning with low latency, supporting a 1M token context and multimodal inputs (audio, video, images).

Flagship frontier model for high-precision multimodal reasoning. Combines strong performance across text, code, and media with a 1M-token context window. Excels at complex agentic tasks.

Lightweight reasoning model optimized for ultra-low latency and cost efficiency. Offers improved throughput and speed for common tasks compared to earlier Flash models.

Workhorse model designed for advanced reasoning, coding, and scientific tasks. Includes built-in thinking capabilities for accuracy and nuanced context handling.

State-of-the-art model for advanced reasoning, coding, and math. Achieves top-tier performance on benchmarks with enhanced problem-solving abilities.

Economical model offering significantly faster time-to-first-token while maintaining quality comparable to larger models like Gemini Pro 1.5.

Fast and efficient model with enhanced multimodal understanding and coding capabilities. Delivers robust agentic experiences with improved instruction following.

GPT-5.4 Pro is OpenAI’s most advanced reasoning model, optimized for complex and high-stakes tasks. With a massive long-context window and strong instruction following, it excels at agentic coding, deep analysis, and multi-step problem solving.

GPT-5.4 is OpenAI’s latest frontier model, unifying GPT and Codex into a single system. Built for strong reasoning, coding, and long-context workflows, it performs reliably across general tasks, document analysis, and software engineering.

Optimized for software engineering and long, complex coding tasks. Supports project building, debugging, and refactoring with dynamic reasoning effort.

Fast, lightweight model optimized for low-latency chat. Uses adaptive reasoning for harder queries while maintaining a warm, conversational style.

Advanced model with major improvements in agentic coding and long-context performance. Optimized for complex tasks requiring step-by-step reasoning and high accuracy.

Frontier-grade model with strong agentic and long-context performance. Delivers consistent gains across math, coding, and science with adaptive reasoning.

Strong general-purpose reasoning model with improved instruction adherence. Features adaptive reasoning and a natural conversational style.

Lightweight, low-latency model for high-throughput chat. Balances speed with strong general intelligence and adaptive reasoning.

Specialized for coding workflows. Supports interactive development and long execution tasks like refactoring and code review.

Smaller, faster version of GPT-5.1-Codex, optimized for efficiency in coding tasks.

Designed for advanced, natural, and context-aware conversations. specialized for enterprise applications with multimodal support.

Advanced model offering improvements in reasoning and code quality. Optimized for complex tasks with step-by-step reasoning.

Compact version of GPT-5 for lighter-weight reasoning tasks. Provides instruction-following benefits with reduced latency and cost.

Smallest and fastest GPT-5 variant. Optimized for developer tools and rapid interactions requiring ultra-low latency.

Open-weight 117B MoE model designed for high-reasoning and agentic use cases. Supports configurable reasoning depth and native tool use.

Open-weight 21B MoE model optimized for low-latency inference. Supports reasoning configuration and agentic capabilities like function calling.

Flagship model optimized for advanced instruction following and software engineering. Features a 1M token context and outperforms GPT-4o.

Mid-sized model competitive with GPT-4o at lower cost. Strong coding and vision capabilities with a 1M token context.

Fastest and cheapest GPT-4.1 model. Exceptional performance for its size, ideal for classification and autocompletion with 1M context.

Advanced small model. More affordable than previous frontier models while maintaining SOTA intelligence and multimodal capabilities.

Omni model supporting text and image inputs. Twice as fast and 50% cheaper than GPT-4 Turbo with improved multilingual and visual performance.

Most capable Sonnet model with frontier performance in coding and agents. Excels at complex codebase navigation and project management.

Strongest model for coding and long-running professional tasks. Designed for agents operating across entire workflows with deep contextual understanding.

Frontier reasoning model optimized for complex software engineering and agentic workflows. Features strong multimodal capabilities.

Fastest and most efficient Claude model. Delivers near-frontier intelligence with extended thinking for controllable reasoning depth.

Optimized for real-world agents and coding workflows. State-of-the-art coding performance designed for extended autonomous operation.

Enhances Sonnet 3.7 capabilities. Excels in coding and reasoning with improved precision and controllability.

Advanced model with improved reasoning and coding. Introduces a hybrid reasoning approach for choosing between rapid responses and extended processing.

High-efficiency model with strong reasoning and agentic tool-use. Uses sparse attention for long-context scenarios.

Hybrid reasoning model supporting thinking and non-thinking modes. Comparable to DeepSeek-R1 on difficult benchmarks but faster.

High-performance open-weight reasoning model optimized for math, coding, and structured problem-solving. Delivers strong chain-of-thought reasoning, competitive benchmark results, and efficient inference, making it well-suited for advanced agents, technical analysis, and logic-heavy workflows.

685B Mixture-of-Experts model. Strong performance across a variety of tasks, serving as a predecessor to V3.1.

70B instruction-tuned model optimized for multilingual dialogue. Outperforms many open and closed models on industry benchmarks.

Fast and efficient 8B instruct-tuned model. Strong performance for its size compared to leading closed-source models.

70B instruct-tuned model optimized for high-quality dialogue. Demonstrated strong performance in human evaluations.

24B model optimized for instruction following and function calling. Significant improvements in coding and STEM benchmarks.

Mistral Small 3. Efficient 24B model for low-latency performance. Competitive with larger models while being faster.

12B parameter model with 128k context. Multilingual support and function calling capabilities.

High-speed native vision-language model built with hybrid linear attention and sparse MoE architecture for efficient inference. Delivers major performance gains over the 3 series across text and multimodal tasks, balancing fast responses with strong overall capability.

MoE code generation model optimized for agentic coding. Features function calling and long-context reasoning over repositories.

Multilingual MoE model optimized for general text, math, and code. Delivers significant gains in long-context reasoning.

Dense 32B model optimized for reasoning and dialogue. Supports seamless switching between thinking and non-thinking modes.

Improved 7B model with better coding and math capabilities. Supports long context up to 128k tokens and multilingual output.

Advanced agentic search system. Designed for deeper reasoning and analysis, planning and executing entire research workflows.

Premier reasoning model powered by DeepSeek R1. Supports in-depth, multi-step queries with larger context and more citations.

Enterprise-grade model for in-depth, multi-step queries. Handles longer and more nuanced searches with added extensibility.

Research-focused model for multi-step retrieval and synthesis. Autonomously searches and evaluates sources for comprehensive reports.

Lightweight, fast, and affordable model. Optimized for speed and simple question-and-answer features with citations.

Agentic tool calling model. Shines in real-world use cases like customer support and deep research with a 2M context window.

Multimodal model with SOTA cost-efficiency. Features a 2M token context window and comes in reasoning and non-reasoning flavors.

Grok Code Fast 1 is a speedy and economical reasoning model that excels at agentic coding. With reasoning traces visible in the response, developers can steer Grok Code for high-quality work flows.

Advanced multimodal reasoning model with a 256k token context window. Supports parallel tool calling, structured outputs, and both text and image inputs. Reasoning is always enabled and cannot be disabled or adjusted. Pricing increases for requests exceeding 128k total tokens, making context management important for cost control.

A lightweight model that thinks before responding. Fast, smart, and great for logic-based tasks that do not require deep domain knowledge. The raw thinking traces are accessible.

Native multimodal model with state-of-the-art visual coding capabilities. Delivers strong performance in general reasoning.

Advanced open reasoning model optimized for persistent step-by-step thought. Supports complex reasoning workflows spanning hundreds of turns.

Large-scale MoE model with 256k context. Improved agentic coding and frontend coding with better generalization.

Mercury 2 is an ultra-fast reasoning diffusion LLM designed for latency-sensitive workloads. By generating and refining tokens in parallel, it delivers extremely high throughput for coding agents, real-time search, and tool-driven workflows.

Flagship open-source foundation model built for complex system design and long-horizon agent workflows. Delivers production-grade performance on large programming tasks, with advanced planning, deep backend reasoning, and iterative self-correction—enabling full system construction and autonomous execution.

Speed-efficient reasoning model built on sparse MoE architecture. Capable of handling long contexts effectively.

Frontier-scale open-weight 400B MoE model (13B active per token) built for creative writing, role-play, chat, and real-time voice. Supports agentic workflows, complex toolchains, and long constraint-heavy prompts, with up to 512k context (128k in Preview). Designed for efficient, production-ready deployment with permissive licensing.

Powerful MoE model delivering exceptional performance. Optimized for Korean, English, and Japanese with high efficiency.

Lightweight reasoning-focused model optimized for agentic tasks and RAG. Runs comfortably on edge devices with 32K context.

Compact, high-performance instruction-tuned model. Delivers strong chat quality and efficient edge inference.

Lightweight agent-centric model. Supports hybrid inference modes (thinking vs non-thinking) for versatile interactions.