One API for every video, image, and coding model. Built for production teams.

Z.AI GLM-5.3 flagship reasoning model for complex software engineering and long-horizon agent tasks, with a 1M-token context window.

GPT-6.1 Terra supports visual reasoning, computer use, and writing with text and image input. It has a 1.05M-token context window and up to 128K output tokens. Tool calling uses the Responses API.

Kimi K3.1 flagship with native vision, a 1M-token context window, and strong coding, financial research, modeling, and long-horizon agent capabilities.

MiniMax M3.1 Flash: multimodal reasoning, coding and tool use with a 1M-token context window and adjustable reasoning effort.

DeepSeek-V4.1-Flash-1017 improves benchmark performance while reducing task time and output length, with unchanged API and pricing. It natively supports a 1M context window, vision, reasoning, and tool use.

Google Gemini 4 Argon: reasoning, multimodal input, tools, structured output, and implicit caching.

Anthropic Claude Haiku 5.5 for fast, high-volume classification, extraction, routing, and subagent tasks.

GPT-6.1 Sol delivers near-Astra performance at a lower cost for complex coding, computer use, and professional work. Compare it with Astra on your tasks to assess the tradeoff between quality and cost. 1.05M context window; reasoning effort from low to max.

Anthropic Claude Sonnet 5.5 for fast, capable coding and everyday knowledge work.

GPT-6 Luna is our most efficient model for focused, high-volume tasks. It is the fastest, most cost-efficient GPT-6 tier—suited for high-volume, latency-sensitive chat, classification, and lightweight agent work; at higher reasoning effort it can handle complex software engineering and computer-use tasks. 1.05M context window; reasoning effort from none to max.

Z.AI GLM-5.3-FlashX is the high-speed native multimodal model for coding, agents, and professional workflows, with a 1M-token context window.

GPT-6 Astra is our most capable model, built for the hardest end-to-end work. Use it for complex reasoning, coding, computer use, research, and document creation. 1.05M context window; reasoning effort from low to max.

Google Gemini 3.8 Flash, configured for reasoning, multimodal, tool, structured-output, and cache workloads.

Anthropic Claude Fable 5.1 for demanding reasoning, long-horizon agentic coding, multistep research, and complex knowledge work.

Tencent Hunyuan Hy4 Preview is an open-source hybrid-reasoning MoE model for coding, agentic tasks, tool use, and long-context workloads.

Z.AI GLM-5.3-Flash is a cost-efficient native multimodal model for coding, agents, and professional workflows, with a 1M-token context window.

Qwen3.8 Flash is Qwen's cost-efficient multimodal reasoning model with a 1M-token context window, 131K maximum output, tool calling, structured output, and native Chat Completions, Responses, and Messages support.

DeepSeek V4 Pro official GA release (0813). Supports thinking mode; up to 1M context tokens.

Google Gemini 3.5 Flash-Lite, generally available July 21, 2026.

Qwen3.8 Max is Qwen's flagship multimodal reasoning model with a 1M-token context window, 131K maximum output, tool calling, structured output, and optional thinking.

ByteDance Dola Seed 2.1 Pro multimodal reasoning model for complex agentic, coding, and visual understanding tasks

Anthropic's flagship model for long-horizon agentic coding.

Qwen3.6-Plus is the balanced mid-tier model in Alibaba's Qwen3.6 family, released around April 2026 and positioned for real-world enterprise agent workflows. Reported features include a 1M-token context window, multimodal vision input and a hybrid architecture combining linear attention with sparse mixture-of-experts routing for inference efficiency. It sits between the speed-first Flash tier and the top Max-Preview tier, targeting teams that need strong reasoning and agentic coding without flagship cost. Ideal for production agent deployments, document and screenshot analysis and autonomous development tasks.

GLM-5V-Turbo is Z.AI's first native multimodal agent foundation model, released April 1, 2026, built on the GLM-5 base and handling image, video and text inputs in a perceive-plan-execute loop. It uses a CogViT vision encoder with multi-token-prediction inference and 30-plus task joint reinforcement learning, and can generate executable code directly from design mockups and visual inputs. It carries roughly 202K context and 131K max output, offered as a proprietary API model. Ideal for vision-based coding and agentic front-end generation from visual assets.

Our strongest mini model yet for coding and subagents. GPT-5.4 Mini brings the strengths of GPT-5.4 to a faster, more efficient model designed for high-volume workloads. 400K context window; configurable reasoning effort.

Gemini 3.1 Pro Preview is the preview release of Google DeepMind's flagship Gemini 3.1 Pro, launched February 19, 2026 as a Transformer-based Mixture-of-Experts model tuned for deep reasoning, agentic coding and tool use. It is natively multimodal across text, image, video, audio and code, with a roughly 1M-token input context window and up to 65K output tokens, enough for entire codebases, long PDFs, or hours of audio/video in one prompt. It delivers a reported 2x-plus reasoning gain over Gemini 3 Pro and tops many tracked benchmarks, with strong instruction following and long-horizon agentic stability. Best for the most complex reasoning, synthesis and agentic engineering tasks where capability matters more than cost.

Claude Opus 4.6 is an Anthropic flagship model released February 5, 2026, building on Opus 4.5 with higher reliability and precision for coding, agents and enterprise workflows. It plans more carefully, sustains agentic tasks for longer, operates reliably in larger codebases and has stronger code review and debugging skills. The model features a 1M-token context window in beta, with standard pricing held at $5 per million input and $25 per million output tokens. Ideal for large-codebase engineering and sustained enterprise agent workloads.

gpt-oss-20b is OpenAI’s medium-sized open-weight model for low latency, local, or specialized use cases. It has 21B parameters with 3.6B active parameters, a 131,072-token context window, and a knowledge cutoff of June 1, 2024. Released under the Apache 2.0 license. Reasoning effort can be set to low, medium, or high.