AI Model Pricing

OpenAI's maximum-capability GPT-5.6 model for demanding reasoning, coding, and multimodal agent workflows.

The balanced GPT-5.6 tier for everyday agents, with frontier capability at half the Sol token price.

The efficient GPT-5.6 tier for high-volume workflows, classification, drafting, and tool routing.

OpenAI's previous-generation flagship, retained for compatibility with existing agents. GPT-5.6 is recommended for new setups.

Fast and efficient version of GPT-5.4 with vision support. Great for high-throughput workloads and computer use tasks.

Anthropic's current Sonnet. The balanced default for most work, and priced below the Sonnet 4.6 it succeeds.

Anthropic's most capable model. Best for nuanced analysis, agentic coding, and complex multi-step tasks.

The smallest, fastest Claude. Well suited to high-volume work and quick turns where cost matters most.

Anthropic's newest and most capable model. Best for nuanced analysis, agentic coding, and complex multi-step tasks.

Frontier-tier reasoning and coding with deep agentic reliability. A step below Opus 4.8 at the same price point.

Previous-generation Opus flagship. Still excellent for nuanced analysis, creative writing, and complex multi-step tasks.

The best balance of intelligence, speed, and cost. Ideal default for most use cases.

Google's current frontier Flash model, built for long-horizon agent work. Vision and video input, 1M context.

The previous Flash generation, at the same price as 3.8. Vision and video input, 1M context.

Google's frontier reasoning model with enhanced software engineering performance and improved agentic reliability.

Google's newest mid-tier multimodal model. Bigger reasoning step up from Gemini 3 Flash, with vision and video input.

Google's fast and affordable multimodal model. Great for vision tasks, quick reasoning, and high-throughput workloads.

Low-latency, cost-effective subagent option with vision and video input.

Ultra-efficient and fast. Great for high-volume tasks, quick responses, and the most cost-efficient workflows.

The previous-generation GA Flash. Kept selectable for cost-conscious setups already running it.

The cheapest Gemini on this list. Best for very high-volume, simple turns.

xAI's flagship since August 2026. 500K context, built for long-running agents, with X search available as a tool.

The cheapest capable Grok on this list, with a 1M-token context. A good everyday default.

A previous xAI flagship, and cheaper than 4.6 at base rates. Kept for agents already pinned to it.

Ultra-low-cost flash model with 1M context. A strong choice for budget-conscious chat and agent workflows.

DeepSeek's frontier model. 1M context, strong coding and reasoning at a fraction of the cost of other premium models.

Tencent's Hunyuan 3 preview — a remarkably cheap large model with a 256K context. One of the lowest-cost capable models on OpenRouter.

Next-gen multimodal model for long-horizon coding, UI/UX generation, and multi-agent orchestration.

Z.ai's latest coding and agentic model with a 1M context window. Cheaper than GLM 5.1 with a much longer context.

Coding-focused model with 202K context. Strong at long-horizon agentic tasks and tool use.

Self-evolving model with strong coding and reasoning. Affordable pricing with 200K context window.

StepFun's newest flash model with vision and video input. Stronger reasoning than Step 3.5 Flash while staying low-cost.

Sparse MoE with 196B total params, 11B active per token. StepFun's most capable open-source model at an ultra-low price.

Xiaomi's newest flagship, tuned for agentic workflows. Strong coding and orchestration with a 1M context window.

Multimodal MiMo with vision and video input at an ultra-low price. Great for high-volume agent and chat workloads.

Xiaomi's flagship 1T-parameter model optimized for agentic workflows. Strong at coding, reasoning, and complex orchestration tasks.

Hybrid linear-attention + sparse MoE flagship. Strong at agentic coding and front-end tasks (78.8 on SWE-bench Verified) with 1M context and native reasoning mode.

Open-weight multimodal Gemma model with vision and video input. MoE architecture for efficient inference.

Larger Gemma variant with vision and video support. Great balance of quality and cost for multimodal tasks.

Auto-routes each request to a working free model on OpenRouter. When one free model goes dark, the router picks another — the most reliable way to stay on free tier.

Hybrid Mamba-Transformer MoE model — 120B params, 12B active. 1M context window with zero cost. Great for complex agent tasks.

OpenAI's flagship frontier model — their most capable for agentic work, coding, and hard reasoning. 272k context on the subscription route.

Balanced GPT-5.6 tier for everyday work — strong reasoning at lower latency than Sol.

Fast, lighter GPT-5.6 tier for quick replies and high-volume chats.

Fast web search with citations. Good for quick factual lookups.

GPT-5.4 reasoning paired with GPT Image 2 generation. Best-in-class prompt adherence and text rendering.

Fast Nano-Banana-class image generation and editing. Strong at photoreal output and low-latency workflows.

ByteDance Seed's latest text-to-image model. Sharp detail, native Chinese prompt support, and competitive per-image pricing.