Models

Anthropic's smallest and fastest model in the Claude 4.5 lineup, optimized for low cost, low latency, and real-time responsiveness, while delivering near-frontier performance on coding, agentic workflows, and extended reasoning tasks.

Anthropic's most advanced Claude 4.5 model designed for coding, agentic AI workflows, and complex reasoning.

A balanced, faster sibling to Opus with strong reasoning and creative abilities.

Anthropic's advanced Claude 4.6 Sonnet model optimized for coding, complex reasoning, and agentic task execution.

Anthropic's most advanced flagship model in the Claude 4.5 family.

Anthropic's upgraded flagship Claude model from the 4-series, focused on enhanced coding, reasoning, and agentic performance.

Most capable model of Anthropic —state‑of‑the‑art for coding, extended reasoning, and sustained workflows.

Anthropic's flagship Claude 4.6 Opus model delivering top-tier intelligence for the most demanding tasks.

Anthropic's latest and most capable model with industry-leading intelligence for complex reasoning and autonomous task execution.

DeepSeek's fast and ultra cost-efficient flagship with 1M context and optional thinking mode — the most affordable frontier-class model available.

DeepSeek's most capable V4 model with frontier-level reasoning, 1M context, and deep optimization for agentic coding workflows.

Google's most cost-efficient multimodal model in the Gemini 3.1 family, offering the fastest performance for high-frequency, lightweight tasks with a 1-million-token context window and native thinking support.

Ultra-fast lightweight preview focused on speed for simple tasks.

Google DeepMind's next-generation fast and efficient variant of the Gemini 3 family.

Google's fast and cost-efficient reasoning model with a 1-million-token context window, native multimodal support, and adaptive thinking for a wide range of tasks.

Powerful multimodal reasoning model with excellent STEM capabilities.

Refined Gemini 3 Pro with better thinking, improved token efficiency, and enhanced factual consistency. Optimized for software engineering behavior, agentic workflows, and precise multi-step execution across real-world domains.

Lightweight LLaMA optimized for speed and cost.

Ultra-strong LLaMA model with remarkable reasoning.

MiniMax M2.5 available free via OpenRouter — SOTA real-world productivity model scoring 80.2% on SWE-Bench Verified. Fluent in office automation, agentic workflows, and cross-software coordination.

MiniMax latest flagship agentic model designed for autonomous real-world productivity, featuring multi-agent collaboration, live debugging, financial modelling, and full document generation across Word, Excel, and PowerPoint.

Small, dense multimodal model designed for efficient text and vision understanding with deployment-friendly performance.

Mistral Small 4 unifies reasoning, multimodal understanding, and agentic coding capabilities into a single 24B model — the most capable Mistral Small yet.

State-of-the-art open-weight multimodal Mixture-of-Experts (MoE) LLM with enormous context support, strong reasoning, and enterprise-grade performance.

Open-weight small reasoning LLM from Mistral AI.

Frontier-class multimodal model from Mistral AI.

Mistral AI's frontier-class multimodal reasoning model.

Moonshot AI next-generation multimodal model for long-horizon coding, UI/UX generation, and multi-agent orchestration.

Moonshot AI advanced reasoning model with long-horizon thinking.

OpenAI's smallest and most cost-efficient GPT-5.4-class model, optimized for high-volume simple tasks with native reasoning support and a 400k context window.

OpenAI's smallest and fastest GPT-5 variant, optimized for ultra-low latency, high throughput, and cost-efficient text and basic multimodal tasks.

OpenAI's strongest mini model for coding, computer use, and subagents. Brings the strengths of GPT-5.4 to a faster, more efficient model designed for high-volume workloads.

A compact version of GPT-5, optimized for speed and efficiency while retaining strong reasoning and basic multimodal support.

Compact version of GPT‑4.1 for faster yet capable responses.

A cost-efficient reasoning-optimized variant of OpenAI's o1 family.

OpenAI's next-generation flagship model, offering unparalleled intelligence, reasoning, and multimodal capabilities.

Enhanced reasoning and depth—ideal for heavy-duty logic and research.

OpenAI's latest flagship instruction-following model with a 1-million-token context window, strong coding, vision, and agentic task capabilities.

OpenAI's frontier model for complex professional work. The first mainline reasoning model incorporating GPT-5.3-Codex capabilities, with native computer use, 1M context, and state-of-the-art token efficiency.

OpenAI's newest flagship GPT-5 series model.

OpenAI's newest frontier model representing a new class of intelligence. More intelligent and more token-efficient than GPT-5.4, with native computer use, coding capabilities, and a 1M context window.

Maximum-performance version of GPT-5.5 using more compute to think harder and provide consistently better answers. The most capable and accurate model in the GPT-5.5 family.

Maximum-performance version of GPT-5.4 using more compute to think harder. Designed for the hardest professional tasks requiring the highest accuracy.

OpenAI's most advanced professional variant of the GPT-5.2 family.

OpenAI's advanced high-reasoning version of the GPT-5 series, designed to produce smarter, more precise, and in-depth answers.

OpenAI's professional-grade reasoning model that uses significantly more compute per query to deliver higher accuracy on the most demanding tasks.

Lightweight, fast, and real-time search-enabled model for quick, citation-backed web answers.

Premium reasoning-tier with chain-of-thought logic, large context (127k tokens), and extensive citation.

Expert-level research model that autonomously performs dozens of web searches, synthesizes sources, and generates comprehensive reports (typically 2–4 minutes per task).

Real-time reasoning model with multi-step logic and citation sourcing.

Enhanced version with larger context and richer citation output for in-depth multi-step answers.

Alibaba's ultra cost-efficient Qwen3.5 Flash model — fast inference with strong reasoning and 1M context at minimal cost.

Qwen3.6 Plus builds on a hybrid architecture combining efficient linear attention with sparse MoE routing — major gains in agentic coding, front-end development, and reasoning over the 3.5 series.

High-capability Qwen3.5 native vision-language model with hybrid linear attention and sparse MoE architecture, delivering state-of-the-art performance across reasoning, coding, and multimodal tasks.

xAI's newest flagship model with strong agentic tool calling, instruction following, low hallucination rate, and configurable reasoning.

xAI's high-performance variant of Grok 4.1 optimized for low-latency, tool-calling, and agentic workflows with extremely large context support and multimodal capabilities.

A variant of xAI's Grok 4.1 Fast optimized for low-latency, instantaneous responses by skipping internal reasoning tokens; best for straightforward, high-throughput text completion with huge context windows.

Speed- and cost-optimized variant of xAI's flagship Grok 4 model.

A cost- and speed-optimized variant of xAI's Grok 4 Fast.

xAI’s flagship model—strong coding, reasoning, and conversational skills.

xAI's newest flagship model with industry-leading speed, agentic tool calling, and optional reasoning. Combines low hallucination rate with strict prompt adherence.

xAI's flagship Grok 4.20 in fast non-reasoning mode — same frontier capabilities with lower latency when chain-of-thought is not needed.

Z.ai's flagship open-source foundation model engineered for complex systems design and long-horizon agent workflows. Delivers production-grade performance on large-scale programming tasks with advanced agentic planning and iterative self-correction.

GLM-5.1 delivers a major leap in long-horizon coding capability — capable of working independently for 8+ hours on a single task, autonomously planning, executing, and self-improving to deliver complete engineering-grade results.