GPT-6.2 Astra is OpenAI's most capable model for demanding reasoning and coding, released in November 2026. It improves on GPT-6 Astra in agentic coding, computer use and professional knowledge work, writes shorter answers at the same reasoning effort, and costs 20% less at list price. It keeps a 1.05M-token context window with up to 128K output tokens.
Browse AI models
140 models
GPT-6.2 Sol is OpenAI's everyday model for coding and agents at scale, released in November 2026. It upgrades GPT-6.1 Sol at the same list price, answers faster with fewer output tokens, and comes close to GPT-6.2 Astra on coding benchmarks. It supports a 1.05M-token context window with up to 128K output tokens.
Claude Fable 5.2 is Anthropic's flagship model, released in November 2026. It beats Claude Fable 5.1 on agentic coding, computer use and long-running knowledge work, generates output up to 30% faster, and keeps a 1M-token context window with up to 128K output tokens.
Mistral Large 4 is Mistral AI's flagship reasoning model with a 1M-token context window, built for agents, coding and complex multilingual work, with vision, function tools and structured output.
GPT-6.1 Terra is the mid-tier model of OpenAI's GPT-6.1 series, priced below GPT-6.1 Sol. It handles everyday agent, coding and writing work, reads screenshots and interfaces well for computer use, and keeps a 1.05M-token context window.
Kimi K3.1 Pro is Moonshot AI's flagship open-weight reasoning model and the successor to Kimi K3, built for long-horizon coding, agent workflows and knowledge work, with a 1M-token context window and thinking always on.
Gemini 4 Argon is Google's flagship reasoning model for complex software engineering, advanced math and science, autonomous multi-step agents and long outputs, with a 1M-token context.
GPT-6.1 Sol upgrades GPT-6 Sol while remaining below Astra in the GPT-6 family. It supports coding agents, computer interaction, document-intensive assignments, and business processes with multiple stages. The source reports results approaching Astra at a lower cost. Compared with GPT-6 Sol, it improves factual accuracy and adherence to explicit constraints and user intent during agent work.
Claude Sonnet 5.5 succeeds Sonnet 5 for everyday assignments with clear scope. It handles feature development, debugging, and creating documents, presentations, and spreadsheets, with clearer writing than its predecessor. Reasoning stays enabled; adjusting effort balances depth against latency and cost, with lower settings supporting responsive agent workflows.
Claude Opus 5.5 succeeds Opus 5 for demanding coding, reasoning, and professional assignments. It improves repository-wide changes, defect detection, financial and scientific analysis, and interpretation of detailed charts and screenshots. Compared with its predecessor, it uses fewer steps and tokens for comparable work and communicates results more clearly. Adaptive reasoning remains enabled, with effort settings controlling depth, response time, and cost.
GPT-6 Luna is the efficient, fast tier below GPT-6 Sol. It serves conversational applications, classification, and lightweight agents at high volumes or with tight latency requirements. Higher reasoning effort also supports difficult engineering and computer-use assignments. It shares the GPT-6 family’s improvements in factual accuracy and concise communication.
GPT-6 Sol sits between Luna and Astra, providing high-end capabilities at a lower cost than the flagship. It handles professional assignments, coding agents, business process automation, and computer interaction, with particular strengths in sustained work on real repositories. It approaches Astra’s factual reliability and shares its clearer technical communication.
Grok 4.7 succeeds Grok 4.6 as SpaceXAI’s flagship for programming, agents, and professional knowledge work. It improves sustained engineering, checking its own results, managing lengthy context, and preparing documents and presentations. Training emphasizes problems that take many hours and native interaction with the Grok Bot harness. Updated safeguards address jailbreak resistance while supporting legitimate cybersecurity and biology work. Reported benchmarks use xhigh reasoning effort.
Doubao Seed 2.1 Pro is ByteDance's flagship model for production agents, coding and multimodal understanding, built to carry long multi-step tasks to completion.
DeepSeek V4.1 Flash uses a sparse mixture-of-experts design with a Causal Encoder-Decoder architecture. Its 552-billion-parameter backbone activates 8 billion parameters for input processing and 16 billion for output. Native image understanding is trained jointly with text. The model targets coding, terminal and computer-use agents, extended tasks, and long-context analysis. Compressed KV caching reduces cache memory to about one quarter of the previous Flash generation, helping lower the cost of agent workloads.
GPT Image 2.5 Flare generates images. One fixed variant per request.
GPT Image 2.5 Sunburst generates images. One fixed variant per request.
GPT-6 Astra is OpenAI’s flagship for completing demanding professional assignments. Its applications include software engineering, advanced analysis, deep research, scientific work, and preparing documents. It is especially suited to extended agent workflows involving interaction with computers and browsers.
Gemini 3.8 Flash is Google's most intelligent Flash model with significant gains from 3.7 Flash across software engineering, agentic tasks, and multi-step reasoning.
Claude Fable 5.1 improves on Claude Fable 5 across the board, with the biggest gains in agentic coding, long-running agentic workflows and knowledge work such as long code refactors and front-end work.
Qwen3.8 27B is an open-weight dense vision-language model from Qwen.
DeepSeek V4 Pro is a large-scale Mixture-of-Experts model from DeepSeek with 1.6T total parameters and 49B activated parameters, supporting a 1M-token context window.
Gemini 3.7 Flash is Google’s multimodal model for responsive software development and agent workflows. It targets problems requiring several reasoning steps, combining fast interaction with dependable execution across a sequence of actions.
Claude Opus 5 is Anthropic’s flagship for difficult reasoning problems, software development, and extended agent workflows. It supports complete engineering assignments, reviewing code, finding defects, interpreting charts and documents, preparing office deliverables, and coordinating other agents. It follows instructions and uses tools across sustained tasks; lower reasoning effort can help when response time and token usage matter.
Gemini 3.5 Flash Lite is a high-efficiency model from Google with upgraded agentic capabilities. It is suited for subagents that execute focused tasks within complex, multi-agent workflows.
Gemini 3.6 Flash is a high-efficiency model from Google for coding, agentic workflows, and web and app development.
Kimi K3 is Moonshot AI’s open-weight multimodal reasoning model with 2.8 trillion parameters. It targets complex programming, professional knowledge work, and extended agent workflows. Strengths include navigating large repositories, using tools, debugging, and iterating from images, logs, tests, and runtime observations. Its architecture combines KDA with Attention Residuals to improve computational efficiency.
GPT-5.6 Luna prioritizes speed and cost efficiency within the GPT-5.6 family. It is suited to large request volumes and applications requiring quick responses, including conversational interfaces, classification, and lightweight agents. It provides reasoning capabilities for workloads where inference cost is a central constraint.
GPT-5.6 Sol is the flagship of the GPT-5.6 family for difficult reasoning problems and software development. It is particularly suited to terminal-based work, coding assignments with multiple stages, and agent workflows that require sustained problem solving.
GPT-5.6 Terra sits between the Luna and Sol tiers, balancing capability with inference cost. It handles routine programming, reasoning, and agent workflows. The source positions it at approximately half the cost of Sol for applications that need more capability than the fastest tier.
Claude Sonnet 5 supports coding, agent workflows, and professional work with a one-million-token context. It accepts text, images, and files and provides adaptive reasoning with low, medium, high, max, and x-high effort settings. This generation introduces a revised tokenizer and real-time cybersecurity safeguards for certain high-risk activities.
Doubao Seed 2.1 Turbo is ByteDance's fast multimodal model of the Seed 2.1 family for coding and multi-step agent work, with text, image and video input, tool calling and a 256K-token context window.
Seedream 5.0 Pro generates images. One fixed variant per request.
Claude Fable 5 belongs to Anthropic’s Mythos class and handles coding and knowledge work with a high degree of autonomy. It reads text, images, and files, produces text, and supports reasoning over a one-million-token context. It is designed to carry out extended asynchronous assignments, including ambiguous problems with many stages, while checking results and correcting mistakes during execution.
MiniMax-M3 is a multimodal foundation model from MiniMax.
Claude Opus 4.8 supports reasoning with text, images, and files as inputs and text as output, using a one-million-token context. It targets autonomous agents and extended projects that depend on retained context. Applications include complex programming, debugging across multiple stages, project coordination, document preparation, presentations, and data analysis. It is designed to preserve consistency during lengthy asynchronous workflows.
gemini-3-pro-image image generation. Submit a task, check its status and download the result.
gemini-3.1-flash-image image generation. Submit a task, check its status and download the result.
Gemini 3.5 Flash combines efficient inference with coding and reasoning capabilities close to the Pro tier. It accepts text, images, video, audio, and PDFs and is optimized for software development and parallel agent workflows. Thinking defaults to medium effort, with minimal, low, medium, and high settings available to balance depth, speed, and cost.
Mistral Medium 3.5 is Mistral AI's frontier-class reasoning model for agentic coding and enterprise workloads, with vision, function tools and a 256K-token context.