Models & Pricing

Nova 2 Lite is an advanced multimodal reasoning model with 1M context. Dynamically adjusts reasoning depth. Extended thinking on complex problems.

Nova Lite is a multimodal understanding model. Multilingual with reasoning over text, images, and videos. Cost-effective for everyday tasks.

Amazon's fastest and most cost-effective text-only model. Ideal for high-throughput, low-latency tasks.

Amazon's most capable multimodal model for complex reasoning tasks. Best teacher for distilling custom models. Supports text, images, and videos.

Amazon Nova Pro is a multimodal understanding model. Multilingual with reasoning over text, images, and videos.

Lightweight, efficient embedding model for high accuracy retrieval tasks. Supports flexible embedding sizes (1024, 512, 256) and 100+ languages.

Image model used in ChatGPT.

Claude Haiku 4.5 delivers near-frontier performance for a wide range of use cases, and stands out as one of the best coding and agent models–with the right speed and cost to power free products and high-volume user experiences. Use cases: Powering free tier user experiences: Claude Haiku 4.5 delivers near-frontier performance at a cost and speed that makes powering free agent products and agentic use cases economically viable at scale. Real-time experiences: Claude Haiku 4.5's speed is ideal for real-time applications like customer service agents and chatbots where response time is critical. Coding sub-agents: Use Claude Haiku 4.5 to power sub-agents, enabling multi-agent systems that tackle complex refactors, migrations, and large feature builds with quality and speed. Financial sub-agents: Use Claude Haiku 4.5 to monitor thousands of data streams—tracking regulatory changes, market signals, and portfolio risks to preemptively adapt compliance and trading systems at previously impossible scales. Research sub-agents: Perform parallel analyses across multiple data sources while maintaining fast response times. Ideal for rapid business intelligence, competitive analysis, and real-time decision support. Business tasks: Claude Haiku 4.5 is capable of producing and editing office files like slides, documents, and spreadsheets. It also better supports strategy and campaign planning, business analysis and brainstorming.

Claude Opus 4 is Anthropic's most intelligent model and is state-of-the-art for coding and agent capabilities, especially agentic search. It excels for customers needing frontier intelligence: Advanced coding: Independently plan and execute complex development tasks end-to-end. It adapts to your style and maintains high code quality throughout. AI agents: Enable agents to tackle complex, multi-step tasks that require peak accuracy. Agentic search and research: Connect to multiple data sources to synthesize comprehensive insights across repositories. Long-horizon tasks and complex problem solving (virtual collaborator): Unlock new use cases involving long-horizon tasks that require memory, sustained reasoning, and long chains of actions. Content creation: Create human-quality content with natural prose. Produce long-form creative content, technical documentation, marketing copy, and frontend design mockups.

Claude Opus 4.1 is Anthropic's most intelligent model and an industry leader for coding and agent capabilities, especially agentic search. It excels for customers needing frontier intelligence: Advanced coding: Independently plan and execute complex development tasks end-to-end. It adapts to your style, thoughtfully plans and pivots, and maintains high code quality throughout. Long-horizon tasks and complex problem solving (virtual collaborator): Unlock new use cases involving long-horizon tasks that require memory, sustained reasoning, and long chains of actions. AI agents: Enable agents to tackle complex, multi-step tasks that require peak accuracy. Agentic search and research: Connect to multiple data sources to synthesize comprehensive insights across repositories. Content creation: Create human-quality content with natural prose. Produce long-form creative content, technical documentation, marketing copy, and frontend design mockups. Memory and context management: Incorporates memory capabilities that allow it to effectively summarize and reference previous interactions.

The next generation of Anthropic's most intelligent model, Claude Opus 4.5 is an industry leader across coding, agents, computer use, and enterprise workflows. Use cases: Coding: Opus 4.5 can confidently deliver multi-day software development projects in hours, working independently with the technical depth and taste to create efficient and straightforward solutions. It has improved performance across coding languages, with better planning and architecture choices - making it the ideal model for enterprise developers. Agents: Claude Opus 4.5, paired with our advanced tool use capabilities, enables more capable agents with new behaviors. Computer use: Our best computer-using model yet, Claude Opus 4.5 navigates new experiences with confident, consistent approaches that deliver more human-like browsing, enabling better web QA, workflow automation, and advanced user experiences. Enterprise workflows: Opus 4.5 can power agents that manage sprawling professional projects from start to finish. It better leverages memory to maintain context and consistency across files, alongside a step-change improvement in creating spreadsheets, slides, and docs. Financial analysis: Opus 4.5 connects the dots across complex information systems - regulatory filings, market reports, internal data - making sophisticated predictive modeling and proactive compliance possible. Cybersecurity: Opus 4.5 brings professional-grade analysis to security workflows, correlating logs, vulnerability databases, and threat intelligence for proactive threat detection and automated incident response.

Claude Opus 4.6 is the next generation of our most intelligent model, and the world's best model for coding, enterprise agents, and professional work. Use cases include: Agents: Opus 4.6 is the world's best model for agentic workflows, orchestrating complex tasks across dozens of tools with industry-leading reliability. It proactively spins up subagents, parallelizes work, and drives tasks forward with minimal oversight. Coding: Opus 4.6 is the world's best coding model, excelling at long-horizon projects, complex implementations, and large-scale codebases. It handles the full lifecycle from architecture to deployment—so senior engineers can delegate their most complex work with confidence. Enterprise workflows: Opus 4.6 sets the standard for enterprise workflows, powering agents that manage sprawling projects end-to-end with professional polish, domain awareness, and industry-leading performance on spreadsheets, slides, and docs. Financial analysis: Opus 4.6 is Anthropic's most capable model for financial workflows, surfacing insights that would take analysts days to compile. It handles the nuance and precision that compliance-sensitive work demands. Cybersecurity: Opus 4.6 delivers the deepest reasoning for security workflows, catching subtle patterns and complex attack vectors with unmatched accuracy. Computer use: Opus 4.6 is our most capable computer-use model for complex workflows, bringing deep reasoning to multi-step tasks that span multiple applications and require planning and judgment.

Claude Opus 4.7 is Anthropic's most capable model — 13% coding lift over Opus 4.6, tripled image resolution (2576px / 3.75 MP), new xhigh effort level, and task budgets for autonomous agent loops. Best for coding, agents, enterprise workflows, cybersecurity, and financial analysis.

Claude Sonnet 4 balances impressive performance for coding with the right speed and cost for high-volume use cases: Coding: Handle everyday development tasks with enhanced performance-power code reviews, bug fixes, API integrations, and feature development with immediate feedback loops. AI Assistants: Power production-ready assistants for real-time applications—from customer support automation to operational workflows that require both intelligence and speed. Efficient research: Perform focused analysis across multiple data sources while maintaining fast response times. Ideal for rapid business intelligence, competitive analysis, and real-time decision support. Large-scale content: Generate and analyze content at scale with improved quality. Create customer communications, analyze user feedback, and produce marketing materials with the right balance of quality and throughput.

Claude Sonnet 4.5 is our most capable model to date for building real-world agents and handling complex, long-horizon tasks–balancing the right speed and cost for high-volume use cases: Long-running agents: Power production-ready assistants for multi-step, real-time applications—from customer support automation to complex operational workflows that require peak accuracy, intelligence, and speed. Coding: Handle everyday development tasks with enhanced performance––or plan and execute complex software projects spanning hours or days––with the ability to save, maintain, and reference information across multiple sessions. Cybersecurity: Deploy agents that autonomously patch vulnerabilities before exploitation––shifting from reactive detection to proactive defense. Financial analysis: Conduct entry-level financial analysis, deliver advanced predictive analysis, or preemptively develop intelligent risk management strategies that leverage best-in-class domain knowledge. Computer use: Claude Sonnet 4.5 is our most accurate model for computer use, enabling developers to direct Claude to use computers the way people do. Research: Perform focused analysis across multiple data sources, turning expert analysis into final deliverables. Ideal for complex problem solving, rapid business intelligence, and real-time decision support.

Claude Sonnet 4.6 delivers frontier intelligence at scale—built for coding, agents, and enterprise workflows.

Mistral's specialized coding model. Optimized for code generation, completion, and analysis.

Zhipu AI CogVideoX 3 — flagship text/image-to-video generation. Up to 5s or 10s, up to 4K resolution.

Zhipu AI CogVideoX Flash — free-tier text/image-to-video generation.

Zhipu AI CogView 3 Flash — free-tier text-to-image generation for trial and low-volume use.

Zhipu AI CogView 4 — text-to-image generation with strong bilingual prompt understanding.

Multilingual multimodal embedding model capable of transforming images, texts, and interleaved content into vector representations. State-of-the-art performance with byte/binary quantization and matryoshka embeddings for compression.

Cohere's most capable model for complex enterprise tasks, RAG, and multi-step reasoning.

Cohere's large model optimized for RAG and enterprise workflows.

Alibaba's multilingual TTS model with natural prosody and voice cloning support.

DeepSeek R1 (671B total, 37B active MoE) is a reasoning model that uses chain-of-thought with <think> tags to solve complex problems. Excels at math, coding, and scientific reasoning tasks with transparent step-by-step thinking.

DeepSeek V3.1 is a hybrid model supporting both thinking and non-thinking modes. Features enhanced tool calling capabilities for agent-based tasks. Thinking mode maintains answer quality comparable to DeepSeek-R1 with improved response times.

DeepSeek V3.1 Terminus — refined variant of V3.1 optimized for tool calling and structured generation tasks.

DeepSeek V3.2 (685B total, 37B active MoE) harmonizes high computational efficiency with superior reasoning and agent performance. Features DeepSeek Sparse Attention for long-context efficiency and a scalable reinforcement learning framework. Excels at long-context reasoning, tool-using agents, function calling, JSON output, and FIM.

DeepSeek V3.2 Exp — experimental variant of V3.2 with enhanced general-purpose capabilities. Strong at tool use, structured output, and multi-turn conversation.

DeepSeek V4 Flash — fast, cost-efficient model with 1M context window. Supports reasoning, tool calling, and structured output.

DeepSeek V4 Pro — high-capability model with 1M context window. Superior reasoning, coding, and agent performance with tool calling and structured output.

Mistral's specialized coding model (123B parameters). Optimized for code generation, analysis, and software engineering tasks.

Doubao 1.5 Lite (32k context) — cost-efficient ByteDance chat model for high-volume routine tasks.

Doubao 1.5 Pro — ByteDance flagship general-purpose chat model with tools and JSON mode.

Doubao 1.5 Vision Pro (32k context) — extended-context vision-language variant.

Doubao Seed 1.6 — ByteDance Seed-series next-gen general model with tools and structured output.

Doubao Seed 1.6 Flash — ultra-low-latency variant of Seed 1.6, ideal for chat and agent loops.

Doubao Seed 1.6 Vision — vision-language Seed 1.6 variant for multimodal understanding.

Doubao Seed 1.8 — incremental upgrade of Seed 1.6 with improved tool-call reliability.

Doubao Seed 2.0 Code — coding-specialized Seed 2.0 model for code generation, refactor, and review.

Doubao Seed 2.0 Lite — cost-efficient Seed 2.0 variant.

Doubao Seed 2.0 Mini — smallest Seed 2.0 variant for high-QPS edge use cases.

Doubao Seed 2.0 Pro — flagship Seed 2.0 model with strongest reasoning and tool use.

Doubao Seed Character — roleplay / persona-driven chat model.

Doubao Seed Code — code generation and code understanding model.

Doubao SeedDance 2.0 — text/image-to-video generation, flagship quality tier.

Doubao SeedDance 2.0 Fast — faster, lower-cost variant of SeedDance 2.0 for iterative video drafting.

Doubao SeedDream 4.5 — text/image-to-image generation, Chinese-bilingual prompt support.

Doubao SeedDream 5.0 — latest text/image-to-image generation with improved fidelity.

Zhipu AI Embedding 2 — text embedding model with fixed 1024-dim output.

Zhipu AI Embedding 3 — latest text embedding model. Default 2048-dim, supports custom dimensions (256, 512, 1024, 2048).

Baidu ERNIE 4.5 (300B MoE, 47B active) — Baidu's flagship model with strong Chinese language understanding.

Fish Audio's TTS model optimized for Chinese and English speech synthesis.

Open-weight 12B variant of FLUX.1 Kontext. Cheapest entry point for image editing. Routed via SiliconFlow.

Highest-quality FLUX.1 Kontext variant. Slower than Pro but yields the best edit fidelity. Routed via SiliconFlow.

Black Forest Labs' image-edit model. 12B parameters, flow-matching diffusion transformer. Edits an input image based on a text instruction while preserving composition. Routed via SiliconFlow.

Workhorse model for all daily tasks. Strong overall performance and low latency supports real-time applications. Suitable for chat interactions, content generation, and general-purpose AI tasks.

Google's cost-effective Gemini model to support high throughput. Optimized for the most price-sensitive use cases while maintaining solid quality for everyday tasks.

Best for balancing reasoning and speed. Gemini 2.5 Flash offers thinking capabilities with strong performance across coding, math, and reasoning tasks at an efficient price point.

Most balanced Gemini model for low latency use cases. Optimized for high-volume, cost-sensitive workloads with strong quality at minimal cost.

Strongest Gemini model quality, especially for code and complex prompts. Features advanced reasoning with thinking capabilities and excels at multi-step problem solving, code generation, and mathematical reasoning.

Google's Gemma 2 9B instruction-tuned model. Lightweight and efficient for basic tasks.

Google's open-source Gemma 3 12B model with vision. Efficient and fast for everyday tasks.

Google's open-source Gemma 3 27B model. Strong performance with vision capabilities in a compact package.

Google's smallest Gemma 3 model at 4B parameters. Lightweight chat, copilots, coding and reasoning, cost-effective fine-tuned vertical assistants.

Google's Gemma 4 26B Mixture-of-Experts model with 4B active parameters per token — open weights under Apache 2.0. Ranks #6 on the open Arena leaderboard. Multimodal text + image input.

Google's Gemma 4 31B dense model — open weights under Apache 2.0, ranks #3 on the open Arena leaderboard. Multimodal text + image input. Released April 2026.

Zhipu AI GLM 4.1V Thinking Flash — free-tier always-on chain-of-thought vision model supporting image and video inputs.

Zhipu AI GLM 4.1V Thinking FlashX — always-on chain-of-thought vision model supporting image and video inputs for complex visual reasoning.

Zhipu AI GLM 4.5 Air — lightweight mixture-of-experts model tuned for agent workloads and high-throughput inference.

Zhipu AI GLM 4.5V — vision-language model supporting image, video, and document understanding. No function-call support.

Zhipu AI GLM 4.6 — mid-range model balancing capability and cost.

Zhipu AI GLM 4.6V — vision-capable mid-range model for image, video, and document understanding with native function calling.

Zhipu AI GLM 4.6V Flash — free-tier vision model supporting image, video, and document understanding.

Zhipu AI GLM 4.6V FlashX — 9B lightweight vision model with function-calling. Supports image, video, and document inputs. Conservative 16K max output.

Zhipu AI GLM 4.7 (358B MoE). Interleaved thinking before every response and tool call. Preserved thinking across multi-turn conversations.

Zhipu AI GLM 4.7 Flash — free-tier lightweight model (30B total, 3B active MoE). Strong reasoning despite small active params. Rate-limited concurrency.

Zhipu AI GLM 4.7 FlashX — high-concurrency paid variant of GLM-4.7 Flash with enhanced throughput.

Zhipu AI's flagship model (754B total, 40B active MoE). DeepSeek Sparse Attention architecture. Strong math/science reasoning.

Zhipu AI GLM 5 Turbo — optimized for sequential task execution with improved continuity. Lower latency than GLM-5 flagship.

Zhipu AI GLM 5.1 — latest flagship model with enhanced reasoning and coding capabilities.

Zhipu AI GLM 5V Turbo — vision-capable model for image understanding and multimodal tasks.

Zhipu AI GLM Image — flagship text-to-image generation model. 2K resolution, strong Chinese typography support.

Zhipu AI GLM OCR (0.9B) — document parser for PDF/image to structured Markdown text extraction.

Zhipu AI Web Search (Pro) — premium multi-engine ZhipuAI self-developed search with lower empty-result rate and higher recall + accuracy than search-std. Returns structured web results with citations; streaming returns a single chunk.

Zhipu AI Web Search routed through Quark — vertical-content focused with precise retrieval against Quark's index. Returns structured web results with citations; streaming returns a single chunk.

Zhipu AI Web Search routed through Sogou — strong vertical coverage of the Tencent ecosystem (news, Penguin Hao, Zhihu) and authoritative for encyclopedia / medical queries. Returns structured web results with citations; streaming returns a single chunk.

Zhipu AI Web Search (Standard) — basic ZhipuAI self-developed search engine, optimised for cost-effective daily-query workloads. Returns structured web results with citations; streaming returns a single chunk.

Previous generation image generation model.

Cost-efficient version of GPT Image 1.

State-of-the-art image generation model.

OpenAI's most advanced image generation model with native reasoning — thinks before drawing. 2K resolution, multi-image consistency, magazine-quality typography, and image editing. Released April 21, 2026.

OpenAI's open-source 120B model with hybrid reasoning, extended thinking, efficient code generation, agentic search, computer use, and tool use capabilities.

OpenAI's open-source 20B model with hybrid reasoning, extended thinking, efficient code generation, agentic search, and tool use. Cost-effective alternative to the 120B variant.

OpenAI's advanced safety reasoning model (120B). Nuanced policy interpretation, multi-turn safety analysis, and justified decisions for content moderation.

OpenAI's safety classification model (20B). Policy reasoning, content filtering, risk analysis, and justification generation.

OpenAI's smartest non-reasoning model. Excels at instruction following and tool calling with broad knowledge across domains. Features a 1M token context window and low latency.

Smaller, faster version of GPT-4.1. Excels at instruction following and tool calling with a 1M token context window and low latency without a reasoning step.

Fastest, most cost-efficient version of GPT-4.1. Excels at instruction following and tool calling with a 1M token context window and minimal latency.

OpenAI's versatile, high-intelligence flagship model. Accepts text and image inputs, produces text outputs including structured outputs. Best model for most tasks outside reasoning-heavy use cases.

GPT-4o model capable of audio inputs and outputs.

Fast, affordable small model for focused tasks. Accepts text and image inputs, produces text outputs. Ideal for fine-tuning and cost-efficient workloads.

Smaller audio-capable GPT-4o model.

Smaller realtime model for text and audio workflows.

Speech-to-text model powered by GPT-4o mini.

Text-to-speech model powered by GPT-4o mini.

Realtime text and audio model from the GPT-4o family.

Speech-to-text model powered by GPT-4o.

Transcription model that identifies who is speaking when.

OpenAI's intelligent reasoning model for coding and agentic tasks with configurable reasoning effort. Features a 400K context window and 128K max output.

A faster, cost-efficient version of GPT-5 for well-defined tasks. Features reasoning token support with a 400K context window and 128K max output at a fraction of the cost.

Fastest, most cost-efficient version of GPT-5. Great for summarization and classification tasks with reasoning token support. Features a 400K context window and 128K max output.

Version of GPT-5 that produces smarter and more precise responses with deeper reasoning.

Version of GPT-5 optimized for agentic coding in Codex.

OpenAI's previous flagship reasoning model for coding and agentic tasks with configurable reasoning effort. Features a 400K context window and 128K max output.

Version of GPT-5.1 optimized for agentic coding in Codex.

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

Version of GPT-5.1 Codex optimized for long-running tasks.

OpenAI's best model for coding and agentic tasks across industries. Features a 400K context window with 128K max output, reasoning token support, and state-of-the-art long-context reasoning.

Intelligent coding model optimized for long-horizon, agentic coding tasks.

Most capable agentic coding model to date.

Best intelligence at scale for agentic, coding, and professional workflows.

Strongest mini model for coding, computer use, and subagents. Fast and cost-efficient with reasoning token support, 400K context window and 128K max output.

Smallest and fastest GPT-5.4 variant for lightweight agentic tasks. 400K context window and 128K max output with reasoning support.

Version of GPT-5.4 that produces smarter and more precise responses.

Next-generation frontier model with 1M context, advanced reasoning, and multimodal input for agentic, coding, and professional workflows.

Version of GPT-5.5 that produces smarter and more precise responses with enhanced reasoning depth.

Audio inputs and outputs with the Chat Completions API.

Best voice model for audio in, audio out with Chat Completions.

Cost-efficient version of GPT Audio.

Model capable of realtime text and audio inputs and outputs.

Best voice model for audio in, audio out.

Successor to gpt-realtime-1.5 with improved voice synthesis and lower latency. Realtime audio in, audio out.

Cost-efficient version of GPT Realtime.

Realtime speech-to-text translation. Accepts audio in any supported source language, returns translated transcript in target language (typically English).

Whisper-class automatic speech recognition over the realtime channel. Audio in, transcribed text out.

xAI's previous flagship model with 131K context window. Strong general-purpose performance with function calling and structured output support.

Cost-efficient reasoning model from xAI with 131K context window. Ideal for tasks requiring reasoning at lower cost with function calling and structured output support.

xAI's most powerful reasoning model with 256K token context window. Excels at complex reasoning, coding, and multi-step problem solving with function calling and structured outputs.

xAI's fast model with 2M token context window. Optimized for speed without reasoning overhead, supporting text and image inputs with function calling and structured outputs.

xAI's fast reasoning model with 2M token context window. Combines speed with strong reasoning capabilities, supporting text and image inputs with function calling and structured outputs.

xAI's fastest model with 2M token context window. Optimized for speed without reasoning overhead, supporting text and image inputs with function calling and structured outputs.

xAI's latest fast reasoning model with 2M token context window. Combines speed with strong reasoning capabilities, supporting text and image inputs with function calling and structured outputs.

xAI's flagship model (March 2026) with 2M token context window. Fast general-purpose mode without reasoning overhead, supporting text and image inputs with function calling and structured outputs.

xAI's flagship model optimized for multi-agent orchestration with 2M token context window. Designed for agent-to-agent coordination, delegation, and parallel task execution.

xAI's flagship model (March 2026) with 2M token context window and deep reasoning. Best for complex multi-step tasks, analysis, and research with function calling and structured outputs.

xAI's most advanced flagship model — industry-leading non-hallucination rate, agentic tool calling, and instruction following. 1M token context, text + image input, configurable reasoning (none/low/medium/high), function calling, structured outputs.

xAI's specialized coding model with reasoning capabilities. Optimized for code generation, analysis, and debugging tasks with function calling and structured outputs.

Tencent Hunyuan A13B — efficient MoE model for general-purpose chat and tool use.

IndexTeam's neural TTS model with low latency and high quality.

Moonshot AI Kimi K2 Instruct — fast non-thinking variant for efficient chat and tool use.

Moonshot AI's deep reasoning model (1T total, 32B active MoE). Specialist for 200-300 step stable tool orchestration, long-horizon planning, and complex coding. Text-only.

Moonshot AI's flagship multimodal model (1T total, 32B active MoE, 384 experts). Native vision with MoonViT encoder. Thinking and instant modes with tool-augmented reasoning.

Moonshot AI's 2026-04 flagship — 1T-parameter MoE that ties GPT-5.5 on coding benchmarks. Agent swarm scales to 300 sub-agents and 4000 coordinated steps. Open-weight.

InclusionAI Ling Flash 2.0 — fast inference model for general-purpose tasks.

Meta's largest open model at 405B parameters. Frontier-class performance across coding, math, reasoning, and multilingual tasks with 128K context.

Llama 3.1 70B with expanded 128K context, multilinguality, and improved reasoning. Optimized for multilingual dialogue and assistant-like chat.

Llama 3.1 8B with 128K context length, multilinguality, and improved reasoning. Optimized for multilingual dialogue, efficient inference on consumer hardware.

Llama 3.2 11B with vision capabilities. Efficient multimodal model for image understanding at low cost.

Llama 3.2 1B lightweight model with on-device processing for improved security and privacy. Ideal for multilingual dialogue, personal information management, knowledge retrieval, and rewriting tasks on edge devices.

Llama 3.2 3B lightweight model. Delivers highly accurate results with capabilities including text generation, summarization, sentiment analysis, and contextual understanding. Ideal for edge devices and mobile AI.

Llama 3.2 90B with vision capabilities. Strong multimodal performance for image understanding and text generation tasks.

Llama 3.3 70B instruct model delivers on-par performance with the 405B model at lower cost. Optimized for multilingual dialogue with strong reasoning capabilities.

Meta's Llama 3.3 70B tuned for versatile general-purpose tasks via Groq LPU inference.

Llama 4 Maverick (400B total, 17B active, 128 experts MoE) offers industry-leading performance in image and text understanding with support for 12 languages. Great for precise image understanding and creative writing. Our product workhorse model for general assistant and chat use cases.

Llama 4 Scout is a general purpose model with 17B active parameters, 16 experts, and 109B total parameters. Features an industry-leading 10M token context length, enabling multi-document summarization, parsing extensive user activity, and reasoning over vast codebases.

Mistral's reasoning-enhanced small model (24B parameters) with vision capabilities. Uses [THINK]/[/THINK] tags for reasoning. Balances reasoning depth with cost efficiency.

MiniMax M2 is a MoE model blending frontier-level intelligence with efficient active parameters. Engineered for AI agents with strong reasoning, coding, and multilingual performance. Ideal for general-purpose chat/coding, tool use, and high-throughput inference.

MiniMax M2.1 is an open-weight model focused on coding, tool use, and long-horizon task planning. Trained with emphasis on practical benchmarks covering front-end, backend, and workflow automation. General-purpose backbone for agent-based applications with reliable instruction following.

MiniMax M2.1 Highspeed variant with ~100 tokens/sec output speed. Same capabilities as M2.1 at 2x cost for latency-sensitive applications.

MiniMax M2.5 is an agent-native frontier model trained to reason efficiently, decompose tasks optimally, and complete complex workflows under real-world constraints. Combines high inference throughput with RL-focused token-efficient reasoning. Suited for full-stack software projects, research workflows, long-horizon planning, and multi-tool orchestration.

MiniMax M2.5 Highspeed variant with ~100 tokens/sec output speed. Same capabilities as M2.5 at 2x cost for latency-sensitive applications.

MiniMax M2.7 is a frontier reasoning model with interleaved thinking chains and multi-tool orchestration. 204K context with strong performance on agentic workflows, coding, and complex multi-step reasoning tasks.

MiniMax M2.7 Highspeed variant with ~100 tokens/sec output speed. Same capabilities as M2.7 at 2x cost for latency-sensitive applications.

Mistral's efficient 14B parameter model with vision support. Good balance of capability and speed for everyday tasks.

Mistral's smallest model at 3B parameters. Ultra-fast and cost-efficient for lightweight tasks.

Mistral's small 8B parameter model with vision. Cost-effective for simpler tasks and high-throughput workloads.

Mistral's flagship large model. Top-tier reasoning, coding, and multilingual with vision.

Mistral's flagship 675B parameter model. Top-tier reasoning, coding, and multilingual capabilities with vision support.

Mistral's efficient small model. Low-cost option for simple tasks and high throughput.

Mistral's 2026-03 unified small model (119B MoE, 6B active). Combines Magistral (reasoning), Pixtral (multimodal), and Devstral (agentic coding) capabilities into a single model.

Mistral's Mixtral 8x7B mixture-of-experts model. Cost-effective for general tasks.

NVIDIA's efficient hybrid model (30B total, 3.5B active MoE). Mamba-2 + Attention layers with 1M context for edge deployment.

NVIDIA's hybrid LatentMoE model (120B total, 12B active). Mamba-2 + Attention + MoE architecture with 1M context. Multi-Token Prediction for fast inference.

OpenAI's powerful reasoning model that pushes the frontier across coding, math, science, and visual perception. Excels in complex queries requiring multi-faceted analysis. Succeeded by GPT-5.