Model Square

A free automatic-routing model option that selects an appropriate backend model based on the request. Lets users access capable AI without selecting a specific model, useful for testing, experimentation, and default deployments.

No description available.

Anthropic's advanced Claude model optimized for complex reasoning, long documents, and careful instruction following. Strong at coding, analysis, and nuanced writing, ideal for professional and enterprise workloads.

The latest Anthropic flagship combining deep reasoning with strong coding and creative capability. Handles extended context and multi-step tasks with high reliability, suited to sophisticated agents and critical applications.

Anthropic flagship model delivering state-of-the-art deep reasoning, advanced coding, and complex agentic autonomy. Features superior instruction following, 40% reduced per-task operational costs over prior generations, and industry-leading reliability on multi-step enterprise workflows.

Anthropic's balanced model offering high capability with efficient operation. Performs well on reasoning, writing, and coding while keeping costs manageable, a flexible choice for a wide range of applications.

An updated Anthropic model delivering refined reasoning, strong long-context handling, and dependable instruction following. A robust middle-tier option for production assistants and document-heavy workflows.

Anthropic high-speed balanced flagship engineered for agentic software engineering and scalable production workloads. Delivers 30%+ faster output generation than Sonnet 5, enterprise-grade cybersecurity capabilities, and exceptional reasoning efficiency at optimal cost.

A fast and cost-effective language model optimized for speed and efficiency. Delivers solid reasoning and coding capabilities at lower latency and cost, making it ideal for high-throughput applications, real-time assistants, and batch processing workloads.

A fast DeepSeek variant optimized for low-latency inference while retaining strong reasoning and coding ability. Great for interactive chat, agent loops, and cost-effective high-volume workloads where quick responses matter.

A vision-capable DeepSeek model that combines language understanding with image input support. Can describe images, answer questions about visual content, and perform OCR-assisted tasks, ideal for multimodal assistants and document processing.

A high-performance language model optimized for complex reasoning, advanced coding, and mathematical problem solving. Excels at multi-step reasoning, code generation and debugging, and technical documentation. Suitable for developer tools, research assistance, and tasks requiring deep analytical thinking.

DeepSeek's high-capability reasoning model built for demanding tasks such as complex mathematics, algorithm design, and deep code analysis. Delivers stronger output quality at the cost of higher compute, suited to expert workloads.

A 552B-parameter mixture-of-experts model with native multimodal understanding and a 1M-token context window. Uses an asymmetric Causal Encoder-Decoder architecture that activates only 8B parameters per input token and 16B per output token for exceptional inference efficiency. Outperforms several flagship models on agentic and coding benchmarks at significantly lower cost, ideal for high-volume coding, reasoning, and agent-driven applications.

A text-to-speech model converting written text into natural, expressive spoken audio. Supports multiple voices and languages, suitable for voice assistants, audiobook narration, and accessibility applications.

The latest fast Gemini release offering improved reasoning and responsiveness. Ideal for dynamic chat experiences, agent tool calls, and production services that require both speed and dependable quality.

Google Gemini 3.8 Flash is a next-generation Flash model for software engineering, agentic knowledge workflows, and complex reasoning. It accepts text, images, audio, and video, supports a 1M-token context window and 64K-token text output, and offers adjustable effort levels for balancing quality, cost, and latency. It is designed for coding, autonomous agents, multimodal understanding, long documents, computer use, and knowledge-work tasks.

Z.AI's general-purpose language model with balanced performance across reasoning, coding, and natural language understanding. Strong multilingual support and reliable generation make it a solid all-rounder for chat and enterprise applications.

A fast variant of the GLM series designed for low-latency inference. Keeps strong instruction-following and conversational quality while reducing response time, ideal for interactive assistants and high-traffic APIs.

An updated fast GLM model with improved reasoning and efficiency. Well suited to real-time chat, coding assistance, and production deployments that need quick, dependable responses at scale.

A capable general-purpose language model with balanced performance across reasoning, coding, and multilingual tasks. Offers reliable instruction following and consistent output quality for a wide range of applications including assistant workflows, content creation, and knowledge-based Q&A.

Z.AI's flagship reasoning model delivering strong performance on complex problem solving, long-context tasks, and code generation. A high-capability choice for demanding professional and research workflows.

A fast and efficient language model optimized for quick responses and cost-effective inference. Suitable for chat applications, customer support bots, and real-time conversational agents where speed matters. Supports multi-turn conversations with strong instruction-following capabilities.

A frontier large language model from OpenAI designed for complex reasoning, long-context understanding, and high-fidelity instruction following. Excels at multi-step problem solving, code generation, and nuanced dialogue. Suitable for advanced agents, analytics, technical writing, and enterprise applications.

A refined general-purpose language model with improved instruction following, reasoning accuracy, and multilingual fluency. Built for reliable performance across diverse tasks including creative writing, technical analysis, coding assistance, and conversational AI.

OpenAI flagship reasoning model with enhanced instruction following, multilingual fluency, and robust tool-use capabilities. Handles long-form conversations, code, structured data, and creative writing with strong coherence. Best for production assistants, RAG pipelines, and complex workflows.

A specialized variant of the GPT-5 family tuned for speed and efficiency. Delivers strong general language performance with lower latency, making it ideal for interactive chat, real-time assistants, and high-throughput serving.

A top-tier general-purpose language model with strong reasoning, coding, and multilingual capabilities. Optimized for complex task decomposition, creative writing, and professional-grade content generation. Suitable for demanding applications requiring high accuracy and nuanced understanding.

A balanced GPT-5 series model optimized for grounded, factual responses and long-context reliability. Well suited to enterprise Q&A, document analysis, summarization, and data-intensive tasks requiring consistency and accuracy.

OpenAI's most capable frontier model, built for complex reasoning, coding, computer use, research, and document creation. Features a 1M-token context window, adjustable reasoning effort from low to max, and state-of-the-art performance on software engineering, cybersecurity, and scientific discovery benchmarks. Ideal for demanding professional workflows, autonomous multi-step task execution, and applications requiring deep planning and judgment.

The ultra-efficient, latency-optimized tier of the GPT-6 family, designed for high-volume workloads and fast interactive applications. Delivers substantial intelligence and coding performance at a fraction of standard inference costs. Ideal for real-time conversational agents, high-throughput data extraction, automated code review, and responsive developer tooling where speed and budget efficiency are paramount.

An advanced general-purpose reasoning model in the GPT-6 family, engineered for demanding professional workflows, complex software engineering, and multi-step agentic tasks. Trained with the breakthrough alignment and reasoning advances of GPT-6 Astra, it delivers state-of-the-art coding, rigorous factuality, and dependable tool use with exceptional cost efficiency. Well-suited for autonomous agents, deep technical analysis, and enterprise-grade problem solving.

xAI's large model known for strong reasoning, real-time knowledge, and engaging conversational style. Performs well on complex problem solving and up-to-date-world topics, suited to investigative and creative tasks.

A powerful language model with real-time knowledge access and advanced reasoning capabilities. Designed for up-to-date information retrieval, analytical tasks, and creative generation. Strong performance on open-ended questions, brainstorming, and knowledge-intensive workflows.

xAI frontier flagship model engineered for long-horizon software engineering, agentic autonomy, and complex knowledge work. Features a 500K token context window with switchable reasoning effort levels, rigorous self-checking capabilities, and outstanding multi-step task reliability.

Hunyuan's large vision-language model handling both text and image inputs. Strong at visual understanding, image captioning, and multimodal reasoning, useful for content moderation, assistive tools, and visual Q&A.

Hunyuan's open-weight flagship built for real-world agentic work, using a sparse mixture-of-experts design of 770B total parameters with 49B active per token. Handles up to 1,048,576 input tokens and returns up to 64K tokens, accepting text, images, and video natively. Released under Apache 2.0 with BF16 and FP8 checkpoints, it targets long-horizon software engineering, cross-document analysis, game development, and scientific research.

A large-context language model supporting extended input windows for deep document analysis and long-form reasoning. Excels at extracting insights from lengthy texts, cross-document synthesis, and complex multi-step tasks. Ideal for research assistance, legal document review, and technical analysis.

An embedding model for converting text into semantic vector representations. Optimized for retrieval, semantic similarity, and RAG pipelines, enabling accurate search and knowledge-base applications.

Part of the Minimax M2 series, delivering strong reasoning and coding with a 204,800 token context window and up to 131,072 tokens of output per request. Open weights are published under a permissive license, and a highspeed variant is available for workloads that need faster token throughput at a higher per-token price. Suited to long-context reasoning, code generation, and multi-step agent tasks.

A multilingual language model from the Minimax family, optimized for dialogue and conversational AI. Supports natural multi-turn interactions with context retention and persona consistency. Capable of creative writing, role-playing, and customer-facing applications across multiple languages.

A high-quality embedding model from Qwen for converting text into dense vector representations. Supports semantic search, clustering, and retrieval-augmented generation (RAG), with strong multilingual coverage and robust out-of-domain generalization.

A multimodal model from Qwen for text, image, and video understanding with fast response times. It supports a 1M-token context window, a 131K-token maximum output, tool calling, structured output, prefix continuation, and context caching. It is suitable for long documents, codebases, complex conversations, coding assistance, agent workflows, and visual analysis.

A compact 3.8-billion-parameter language model balancing performance and efficiency. Well-suited for resource-constrained deployments, edge inference, and cost-sensitive applications. Offers solid instruction following, reasoning, and multilingual capabilities relative to its size. Good fit for lightweight chatbots, text summarization, and classification tasks.

An agentic model tuned for autonomous, multi-step task execution rather than single-turn answers. It coordinates generation, verification, and iteration as separate stages, so plans are checked and revised before being accepted, and it carries structured memory across a long-running session. Strong on tool invocation, research and discovery pipelines, and algorithm design tasks where results need to survive verification. Well suited to building AI agents for scientific, engineering, and analytical automation.

A production-grade multimodal model delivering high-quality text and image generation. Supports conversational tasks, visual content creation, and creative workflows through a scalable API, ideal for AI-driven products and content platforms.

A next-generation multimodal text model from Agnes AI, designed for agentic coding and tool-driven development workflows. Focuses on stable execution, reliable tool orchestration, long-task context retention, and high-quality output delivery across multi-step agent tasks. Supports text and image input with a 512K context window, making it ideal for building dependable AI agent applications and complex development pipelines.

A production-grade video generation model supporting text-to-video, image-to-video, and keyframe animation workflows. Generates high-quality cinematic video content with strong visual consistency across frames. Supports asynchronous task-based API for scalable video generation. Available in 480p, 720p, and 1080p resolutions with standard aspect ratios including 16:9, 9:16, 1:1, 4:3, and 3:4. Ideal for storytelling, marketing videos, social media content, product demos, and creative AI workflows.

A next-generation video generation model supporting text-to-video, image-to-video, and keyframe animation workflows. Produces cinematic, temporally consistent footage with improved motion realism, sharper detail, and more stable subjects than earlier versions. Offers asynchronous task-based generation across standard aspect ratios including 16:9, 9:16, 1:1, 4:3, and 3:4. Ideal for short films, marketing videos, social media clips, product demos, and creative AI pipelines.

A compact, efficient model designed for responsive everyday AI workloads, including conversation, lightweight reasoning, coding assistance, and structured text generation. It is suited to applications that prioritize low latency, consistent output, and economical inference, such as assistants, automation workflows, and API powered products.

A fast image creation and editing model built for rapid turnaround and high throughput. Generates polished graphics, infographics, and illustrations from text prompts with accurate text rendering, and supports reference-image editing for consistent layouts. Ideal for social media graphics, posters, infographics, marketing assets, and automated design workflows.

A lightweight image creation and editing model optimized for efficiency and cost. Turns text prompts into clean, production-ready graphics and infographics, and supports reference-image editing with flexible subject control. Combines generation and editing in one workflow, ideal for social media graphics, posters, banners, slide visuals, and everyday design content.

A science-oriented multimodal model that pairs broad general-purpose ability with dedicated scientific agents and discovery workflows. Built for long-horizon research spanning disciplines, it handles literature understanding, hypothesis building, solution design, and experimental verification in one continuous process. Suited to chemistry, materials, biology, and earth science work where results must be reproducible and traceable rather than merely plausible.

A streaming speech-to-text model supporting 39 languages including Chinese, English, Japanese, Korean, French, German, Spanish, Italian, Russian, Arabic, Hindi, Thai, Vietnamese, Portuguese, Dutch, and more. Features configurable latency from 80ms to 1s with built-in punctuation and capitalization. Suited for real-time captioning, voice assistants, and multilingual transcription requiring fast, reliable speech recognition.

A mixture-of-experts model built for everyday decision making rather than benchmark scores. Trained on extensive real-world lifestyle and preference data, it targets the fuzzy questions general models handle poorly, such as where to go, what to eat, how to arrange a trip, or how to style a room, with a 512K context window for long reference material. Accepts text, images, and files, and exposes adaptive thinking with controllable reasoning effort. Useful for consumer assistants, recommendation and search-style products, and localized content workflows.

A high-performance large language model delivering fast, accurate responses across conversation, coding, reasoning, and content generation tasks. Balances intelligence with speed and affordability, making it well-suited for chatbots, productivity tools, business automation, and application backends. Accessible through an OpenAI-compatible API.

A versatile image generation model supporting text-to-image and image-to-image transformation. Generates high-quality visuals from natural language prompts with strong prompt adherence and stylistic flexibility. Suitable for creative design, illustration, photo-realistic rendering, and visual content production. Supports multiple resolution presets and aspect ratios for diverse output needs.

An advanced text-to-image generation model designed for high-information-density scenes with complex compositions and rich visual details. Supports text-to-image, image-to-image editing, and multi-image compositing workflows. Optimized for preserving original composition and subject layout during edits. Outputs up to 4K resolution with flexible aspect ratios including 1:1, 3:4, 4:3, 16:9, 9:16, 2:3, 3:2, and 21:9. Ideal for concept art, marketing visuals, product photography, and social media creative assets.

An upgraded text-to-image generation model built for high-information-density scenes with complex compositions and rich visual detail. Supports text-to-image, image-to-image editing, and multi-image compositing, with stronger prompt adherence, finer texture, and improved preservation of composition and subject layout during edits. Outputs up to 4K resolution with flexible aspect ratios including 1:1, 3:4, 4:3, 16:9, 9:16, 2:3, 3:2, and 21:9. Ideal for concept art, marketing visuals, product photography, and social media creative assets.

A real-time multimodal speech model enabling low-latency, streaming voice conversations with natural turn-taking and immediate responses. Supports live dialogue, voice-enabled assistants, and interactive applications that require fluid, human-like interaction. Accessible through an OpenAI-compatible realtime API.

A multifunctional speech understanding model supporting Mandarin Chinese, Cantonese, English, Japanese, and Korean. Performs simultaneous speech recognition, spoken language identification, emotion recognition, and audio event detection (e.g., applause, laughter, crying, coughing) from a single audio input. Non-autoregressive architecture enables exceptionally low inference latency, ideal for real-time voice analytics, content moderation, and multimodal audio understanding.

A high-density agent model built for fast, cost-efficient execution of coding and agent tasks. A sparse mixture-of-experts design activates only a small share of its parameters per token, which keeps time-to-first-token under three seconds and peak output throughput up to 300 tokens per second while reducing iteration steps and token consumption on agent workloads. Reasoning strength is exposed across four controllable levels, with readable reasoning chains at the highest setting. Well suited to tool orchestration, multi-step engineering work, and high-frequency automated pipelines.

A model with one of the largest context windows available, taking 1,000,000 input tokens in a single call and returning up to 524,288 tokens per response. Accepts text, image, audio, and video natively, exposes adjustable reasoning effort, and performs strongly on coding tasks. A practical fit for whole-repository analysis, long document review, and extended agent sessions where the workload must stay in one context instead of being split into many small fragments.

Xfyun (iFlytek) flagship LLM offering strong Chinese-language capability, domain knowledge, and stable generation. Well suited to Chinese enterprise scenarios, customer service, and specialized industry applications.

StepFun's flagship model for agentic work, delivering frontier-level performance on software engineering and professional knowledge work with particular strength in finance. Natively understands text, images, and video within a 1M token context window and returns up to 64K tokens per response, with low, medium, and high reasoning effort available. Built on a sparse mixture-of-experts design with 600B total parameters and 27B active, it suits workloads that need large amounts of input, repeated tool calls, and steady progress toward a finished deliverable.