AI models

Anthropic's most capable publicly available model, built for ambitious long-running agentic work in software engineering, scientific research, and complex knowledge work. Improves on Fable 5 in coding and root-cause debugging, and cuts cache read pricing by 75% to $0.25 per million tokens. Features mandatory adaptive thinking, 1M token context window, 128K output capacity, and all effort levels including xhigh and max. Priced at $10/$50 per million tokens. Knowledge cutoff June 2026.

Previous-generation Fable model for ambitious long-running agentic tasks in software engineering, scientific research, and complex knowledge work. Features mandatory adaptive thinking, 1M token context window, 128K output capacity, and all effort levels including xhigh and max. Anthropic recommends Claude Fable 5.1 for new work.

Anthropic's latest Opus model and the direct successor to Claude Opus 5 at a lower price ($4/$20 per million tokens). Keeps the 1M token context window, 128K output capacity, and the Opus 5 feature set, with adaptive thinking always on and effort control from low through max defaulting to medium. Improves agentic coding and reasoning over Claude Opus 5 and drops cache read pricing to $0.20 per million tokens. Knowledge cutoff June 2026.

Anthropic's latest balanced model and the successor to Claude Sonnet 5 at the same price ($2/$10 per million tokens). Features a 1M token context window billed at standard rates, 128K output capacity, adaptive thinking on by default, and effort control from low through max defaulting to high. Knowledge cutoff June 2026.

Anthropic's latest Opus model for complex agentic coding and enterprise work — approaching Claude Fable 5.1's capability at half the price ($5/$25 per million tokens). Features a 1M token context window, 128K output capacity, adaptive thinking on by default with full effort control (low through max), and stronger self-verification and error recovery so it needs less back-and-forth. Knowledge cutoff May 2026.

Previous-generation Opus model. ~4x more reliable code review, top Super-Agent benchmark (only model to complete every case end-to-end), and 84% on Online-Mind2Web computer use. Features 1M token context window, 128K output capacity, adaptive thinking with effort control, and priority service tier. Anthropic recommends Claude Opus 5 for new work.

Previous-generation Opus model. Features 1M token context window with new tokenizer, 128K output capacity, adaptive thinking (only supported thinking mode), high-resolution image support (2576px / 3.75MP), task budgets, and xhigh effort level. Strong at complex reasoning, agentic coding, knowledge work, and vision tasks; Anthropic recommends Claude Opus 5 for new work.

Anthropic's latest balanced model with adaptive thinking on by default. Features 1M token context window, 128K output capacity, stronger coding and agentic capabilities, and improved vision. Uses a new tokenizer and the effort parameter for reasoning control. Costs $2/$10 per million tokens across the full 1M context.

Anthropic's flagship balanced model with near-Opus level performance at mid-tier pricing. Features 1M token context window, 128K output capacity, exceptional coding capabilities (79.6% on SWE-bench), and advanced computer use (72.5% on OSWorld). Ideal for complex agents, long-context reasoning, and production workflows.

Anthropic's smart model for complex agents and coding tasks. Offers the best balance of intelligence, speed, and cost for most use cases, with exceptional performance in coding and agentic workflows.

Anthropic's most intelligent model with breakthrough capabilities. Features 1M token context window, 128K output capacity, adaptive thinking for optimal reasoning, and agent teams for complex workflows. Best for enterprise agents, production coding, and sophisticated problem-solving.

Anthropic's premium model combining maximum intelligence with practical performance. Delivers state-of-the-art coding capabilities and frontier performance for production code, sophisticated agents, and complex tasks.

Anthropic's fastest model for high-volume work such as classification, extraction, and routing. It has a 1M token context window, supports adaptive thinking with effort control, and costs $0.10/$0.50 per million tokens for prompts up to 100K tokens. Longer prompts use a higher rate.

Anthropic's fastest model with near-frontier intelligence. Matches Sonnet 4's performance on coding, computer use, and agent tasks while being optimized for speed and cost-efficiency.

OpenAI's most capable flagship model, built for the hardest end-to-end work: complex reasoning, coding, research, and document creation. Features a 1.05M token context window, 128K max output, image understanding, and the full reasoning effort range up to max. Knowledge cutoff April 2026. OpenAI documents additional native tools for this family, but qbash currently exposes the verified web search path only.

OpenAI's near-Astra model for complex coding, computer use, and professional work at a lower cost. Features a 1.05M token context window, 128K max output, image understanding, and reasoning effort from low through max, with a default of medium. Knowledge cutoff April 2026. OpenAI documents additional native tools for this model, but qbash currently exposes the verified web search path only.

OpenAI's GPT-6 flagship for complex coding and agentic workflows. Features a 1.05M token context window, 128K max output, image understanding, and the full reasoning effort range up to max, with a default of medium. Knowledge cutoff April 2026. OpenAI documents additional native tools for this family, but qbash currently exposes the verified web search path only.

OpenAI's GPT-6 model for focused, high-volume tasks at scale, optimized for cost. Features a 1.05M token context window, 128K max output, image understanding, and the full reasoning effort range up to max, with a default of medium. Knowledge cutoff May 2026. OpenAI documents additional native tools for this family, but qbash currently exposes the verified web search path only.

OpenAI's flagship frontier model for complex professional work. Official API docs position Sol as the default starting point for complex reasoning and coding. Features a 1.05M token context window, 128K max output, image understanding, and max reasoning effort. The OpenAI API also exposes the alias 'gpt-5.6' for this model family default. OpenAI documents additional native tools for this family, but qbash currently exposes the verified web search path only.

OpenAI's GPT-5.6 model that balances intelligence and cost. Features a 1.05M token context window, 128K max output, image understanding, and max reasoning effort. OpenAI documents additional native tools for this family, but qbash currently exposes the verified web search path only.

OpenAI's GPT-5.6 model optimized for cost-sensitive, high-volume workloads. Features a 1.05M token context window, 128K max output, image understanding, and max reasoning effort. OpenAI documents additional native tools for this family, but qbash currently exposes the verified web search path only.

OpenAI's most capable frontier model. Features 1.05M token context window, native compaction for longer agent trajectories, 33% fewer errors than GPT-5.2, and 18% fewer response-level errors. First mainline model with computer-use capabilities.

OpenAI's maximum-capability model for the most demanding professional workloads. Features 1.05M context window with restricted reasoning options (medium through xhigh) for consistently deep analysis.

OpenAI's first fully retrained base model since GPT-4.5. Features 1.05M token context window, ~40% improved token efficiency, and reasoning defaulted to 'low' (rather than 'none') for stronger out-of-the-box answers. Same reasoning controls as GPT-5.4 — 'none' through 'xhigh' available.

OpenAI's maximum-capability model in the GPT-5.5 family. Features 1.05M token context window with restricted reasoning options (medium through xhigh) for consistently deep analysis. Same parameter shape as GPT-5.4 Pro.

OpenAI's most capable model for professional knowledge work, combining maximum reasoning capabilities with strong performance on complex, multi-step tasks across spreadsheets, presentations, coding, and long contexts.

OpenAI's advanced model for complex knowledge work. Excels at building spreadsheets and presentations, writing code, interpreting images, and working with long contexts. Performs at or above human expert level on well-specified knowledge work tasks.

Fast variant of GPT-5.2 optimized for responsive professional tasks. Delivers GPT-5.2's capabilities with reduced latency for real-time applications.

Fast variant of GPT-5.3 optimized for everyday tasks. Delivers responsive, low-latency performance for real-time applications and conversational workflows.

OpenAI's most capable agentic coding model. Combines Codex and GPT-5 training stacks for best-in-class code generation, reasoning, and general-purpose intelligence. 25% faster than its predecessor. Optimized for long-running agentic coding tasks.

OpenAI's breakthrough model with significant improvements in math, coding, visual perception, and health. 94.6% on AIME 2025, 74.9% on SWE-bench Verified. ~45% less likely to hallucinate than GPT-4o.

Efficient variant of GPT-5 for cost-effective intelligent tasks. Balances strong performance with reduced computational requirements.

Ultra-efficient GPT-5 model optimized for high-volume, cost-sensitive applications. The fastest and most affordable option in the GPT-5 series.

Iterative improvement of GPT-5 with more conversational style, improved instruction following, and adaptive reasoning capabilities.

OpenAI model with major improvements in coding and instruction following. Completes 54.6% of SWE-bench Verified tasks vs 33.2% for GPT-4o. Features 1M token context window and improved long-context comprehension.

Balanced small model matching GPT-4o intelligence at nearly half the latency and 83% lower cost. Features 1M token context window with strong coding and instruction-following performance.

OpenAI's fastest and cheapest model with 1M token context window. Scores 80.1% on MMLU. Optimized for high-volume, low-latency applications requiring basic intelligence.

Fast, affordable small model for focused tasks. Accepts text and image inputs with structured outputs. Optimized for cost-effective intelligent processing.

OpenAI reasoning model trained with reinforcement learning to perform complex reasoning tasks. Designed to think deeply before responding.

OpenAI's most powerful reasoning model. Pushes the frontier across coding, math, science, and visual perception. 20% fewer major errors than o1 on real-world tasks. First reasoning model with agentic tool use including web search and image understanding.

Fast, cost-efficient reasoning model optimized for high-volume STEM tasks. Best-performing benchmarked model on AIME 2024 and 2025. Supports higher usage limits than o3 for throughput-sensitive workloads.

Small reasoning model optimized for science, math, and coding tasks. Offers efficient complex reasoning at lower computational cost.

OpenAI's latest image generation model with improved quality and instruction following.

Google's latest stable Flash model for coding, agents, and complex workflows.

Stable Flash model balancing speed and multimodal reasoning.

Stable low-cost Flash-Lite model for high-volume tasks.

Stable replacement for Gemini 3.1 Flash-Lite Preview.

Google's most capable model for multimodal understanding and advanced agentic tasks. Successor to Gemini 3 Pro Preview with improved reasoning across text, images, video, audio, and PDFs.

Google's most balanced model built for speed, scale, and frontier intelligence. Default model across several Google surfaces with strong coding and state-of-the-art reasoning. Supports thinking, structured outputs, and search grounding across text, images, video, audio, and PDFs.

Google's 2.5 thinking model for complex code, math, and long-context work. Access for new projects is limited.

Google's 2.5 model for large-scale processing and low-latency tasks with thinking capabilities. Access for new projects is limited.

Google's cost-efficient 2.5 Flash-Lite model for high-volume applications. Access for new projects is limited.

xAI's flagship model for coding, agentic tasks, and knowledge work.

xAI's previous frontier model for coding and agentic work.

xAI's coding and agentic software engineering model.

Fast xAI model with strong tool calling and configurable reasoning.

Early-access xAI coding model for agentic software workflows; availability varies by account.

xAI's March 2026 reasoning model for complex tasks and tool calling. Supports X search and web search.

xAI's March 2026 non-reasoning model for fast, direct tasks. Supports X search and web search.

xAI's March 2026 Grok 4.20 variant for multi-agent research and coordination.

xAI's most capable model with X search support

xAI's high-speed model built for low-latency production workloads. Delivers rapid responses while maintaining strong reasoning capabilities. 1M token context window.

xAI's compact, cost-efficient reasoning model. Supports reasoning_effort control for balancing speed vs depth. Ideal for high-volume tasks.

xAI's fastest compact model. Same capabilities as Grok 3 Mini with optimised infrastructure for minimal latency. Supports reasoning_effort control.

xAI's multimodal model with image understanding. Accepts JPEG and PNG images up to 20MB alongside text for richer, context-aware responses.

xAI's image generation model.

xAI's premium image generation model with enhanced quality.

TypeSafe AI's System One decision model. It reads the prompt's text and answers typed questions about it. A pick-one question returns the chosen option with a confidence, a score question returns a value on your scale with a confidence, and a yes-or-no question returns the probability from 0 to 1 that the statement is true. Use it in prompts and tasks to classify or route input.

Xiaomi's MiMo V2.6 Pro via OpenRouter. The flagship MiMo model, at over 1T parameters. Accepts image, audio and video input with a 1M token context window.

Xiaomi's MiMo V2.6 Flash via OpenRouter. An open-source mixture-of-experts model (309B total, 15B active parameters). Accepts image, audio and video input with a 1M token context window.

inclusionAI's Ling 3.0 Flash VL via OpenRouter. A mixture-of-experts model (124B total, 5.5B active parameters) that adds image and video input to Ling 3.0 Flash, with a 262K token context window.

Inception's Mercury 2.5 via OpenRouter. A diffusion reasoning model that generates and refines many tokens in parallel, with a 260K token context window.

Alibaba's Qwen3.8 Max (0902 snapshot) via OpenRouter. A 2.4T-parameter mixture-of-experts reasoning model that accepts image and video input, with a 1M token context window.

Meta's Muse Spark 1.3 via OpenRouter. A multimodal reasoning model for long-running agent and coding work. Accepts image, video and file input with a 1M token context window.

Tencent's Hy4 preview via OpenRouter. A mixture-of-experts model (770B total, 49B active parameters) for coding agents and multi-step tool use, with a 1M token context window.

Z.ai's GLM 5.3 Flash via OpenRouter. A lower-cost multimodal model for coding and agent work, with a 1M token context window.

Alibaba's Qwen3.8 Flash via OpenRouter. A lower-cost multimodal reasoning model for coding and visual analysis, with a 1M token context window.

Z.ai's GLM 5.3 via OpenRouter. A text-only reasoning model for software engineering and long-running agent work, with a 1M token context window.

Alibaba's Qwen3.8 2.4T A95B via OpenRouter. The open-weight release of Qwen3.8 Max, a mixture-of-experts reasoning model (2.4T total, 95B active parameters) with a 1M token context window.

ByteDance Seed's Seed 2.0 Code via OpenRouter. A coding model for frontend and multi-language agent work. Accepts image and video input with a 262K token context window.

Upstage's Solar Pro 4 via OpenRouter. A lower-cost model for long agent tasks and document-heavy office work, with a 524K token context window.

Poolside's Laguna S 2.1 via OpenRouter. A coding agent model (118B total, 8B active parameters) with a 1M token context window.

Meituan's LongCat 2.0 via OpenRouter. A mixture-of-experts model (1.6T total, 48B active parameters) for coding and repository-level changes, with a 1M token context window.

Thinking Machines Lab's Inkling via OpenRouter. An open-weight mixture-of-experts model (975B total, 41B active parameters) for coding and tool use. Accepts image and audio input with a 524K token context window.

Moonshot AI's Kimi K3 via OpenRouter. A 2.8T-parameter open-weight reasoning model for coding and long-running agent work. Accepts image and video input with a 1M token context window.

Tencent's Hy3 via OpenRouter. A mixture-of-experts reasoning model (295B total, 21B active parameters) with configurable reasoning effort and a 262K token context window.

Z.ai's GLM 5.2 via OpenRouter. A text-only reasoning model for long agent workflows and project-level software engineering, with a 1M token context window.

Moonshot AI's Kimi K2.7 Code via OpenRouter. A coding model in the Kimi K2 family for end-to-end programming tasks. Accepts image input with a 262K token context window.

NVIDIA's Nemotron 3 Ultra via OpenRouter. An open reasoning model on a hybrid Transformer-Mamba architecture (550B total, 55B active parameters) with a 262K token context window.

MiniMax's M3 via OpenRouter. A multimodal model for agent work and coding that accepts image and video input, with a 1M token context window.

StepFun's Step 3.7 Flash via OpenRouter. A mixture-of-experts reasoning model with about 11B active parameters. Accepts image and video input with a 262K token context window.

Mistral's Medium 3.5 via OpenRouter. A dense 128B model for agent work and coding that accepts image input, with a 262K token context window.

Moonshot AI's Kimi K2.6 via OpenRouter. A multimodal model for long-running coding work and multi-agent orchestration, with a 262K token context window.

Mistral's Small 4 via OpenRouter. Combines reasoning and image input in one model, with a 262K token context window.

NVIDIA's Nemotron 3 Super via OpenRouter. An open hybrid Mamba-Transformer mixture-of-experts model (120B total, 12B active parameters) for multi-agent work, with a 262K token context window.

Alibaba's Qwen3 Coder Next via OpenRouter. An open-weight coding model (80B total, 3B active parameters) with a 262K token context window.

MoonshotAI's Kimi K2.5 model via OpenRouter. Features massive 262K token context window for handling very long documents and conversations.

Writer's Palmyra X5 model via OpenRouter. Enterprise AI agent model with 1.04M token context window. Features novel transformer architecture and hybrid attention mechanisms for faster inference and expanded memory, optimized for processing large enterprise data volumes.

MiniMax's M2.1 model via OpenRouter. Consistently ranks in top-10 across benchmarks with strong performance in reasoning and coding tasks. Features 196K context window.

Mistral's Devstral 2 model via OpenRouter. 123B-parameter dense transformer model specializing in agentic coding across multiple files. Features 262K context window for handling complex codebases.

Amazon's Nova 2 Lite model via OpenRouter. Multimodal model supporting text, image, video, and file inputs. Excels at document processing, video information extraction, code generation, and multi-step agentic workflows. Features 1M token context window.

Amazon's Nova Premier model via OpenRouter. Flagship AWS model with 1M token context window designed for complex reasoning tasks. Multimodal support for text and image inputs. Can be used as a teacher model for custom distillation.

Mistral AI's mid-tier model via OpenRouter. Balances performance and cost, suitable for a wide range of tasks requiring solid reasoning and generation capabilities.

OpenAI's open-weight gpt-oss-120b via OpenRouter. A 117B-parameter mixture-of-experts reasoning model with 5.1B active parameters and a 131K token context window.

OpenAI's open-weight gpt-oss-20b via OpenRouter. A 21B-parameter mixture-of-experts reasoning model with 3.6B active parameters and a 131K token context window.

Mistral AI's specialized coding model via OpenRouter. Optimized for code generation, completion, and technical tasks with exceptional programming capabilities across multiple languages.

Alibaba's balanced Qwen3 model via OpenRouter with 32B parameters. Apache 2.0 licensed, offering strong performance with efficient resource usage and multilingual support.

Alibaba's efficient Qwen3 model via OpenRouter with 14B parameters. Apache 2.0 licensed, optimized for speed and cost-effectiveness while maintaining quality across multiple languages.

Small reasoning model optimized for science, math, and coding tasks via OpenRouter. Offers efficient complex reasoning at lower computational cost.

Perplexity AI's Sonar model via OpenRouter with native web search integration. Provides real-time information by searching the web and synthesizing current results into responses.

OpenAI reasoning model trained with reinforcement learning via OpenRouter. Designed to think deeply before responding on complex reasoning tasks.

Meta's open-source Llama 3.3 model (70B parameters) via OpenRouter. Instruction-tuned for helpful, high-quality responses.

OpenRouter's intelligent free model router via OpenRouter. Dynamically selects from available free models and smartly filters for required features like image understanding, tool calling, and structured outputs. Perfect for cost-conscious applications.

Relace's agentic search model via OpenRouter. Specialized for multi-step codebase exploration with parallel tool usage. Claims 4x faster performance than frontier models for code search tasks.

MoonshotAI's Kimi K2 with thinking mode via OpenRouter. Trillion-parameter MoE with 32B active parameters supporting persistent multi-turn reasoning. Capable of 200-300+ tool calls with step-by-step thought process.

Perplexity AI's Sonar Pro with advanced search capabilities via OpenRouter. Features autonomous multi-step reasoning for planning and executing entire research workflows. Combines web search with deep reasoning for complex research tasks.

Cohere's premium enterprise model via OpenRouter. Optimized for retrieval-augmented generation (RAG), document processing, and business applications with exceptional instruction-following.

Cohere's balanced enterprise model via OpenRouter. Provides strong performance for RAG and document tasks with efficient resource usage and reliable instruction-following.

Meta's high-performance Llama 3.1 model via OpenRouter with 70B parameters. Balances quality and efficiency with instruction tuning for diverse applications.

Meta's ultra-efficient Llama 3.1 model via OpenRouter with 8B parameters. Optimized for ultra-fast inference and minimal costs while maintaining reliable performance.

Mistral AI's efficient smaller model via OpenRouter. Provides fast inference and lower costs while maintaining good performance for everyday tasks.

Mistral AI's mixture-of-experts model via OpenRouter with 8 experts and 22B parameters each. Delivers excellent quality across diverse tasks with efficient parameter usage.

Mistral AI's flagship model via OpenRouter. Delivers strong performance across reasoning, coding, and multilingual tasks. Provides a compelling European alternative to frontier models.

Perplexity AI's reasoning-enhanced Sonar model via OpenRouter. Combines deep reasoning capabilities with web search integration for complex research and analysis tasks.