Agent chat models: default, Best, and cost
Choose an Agent chat model before you start a thread. This page is the chooser. Open Agent when you are ready to talk.
Our pick
GLM 5.3 FlashOur Pick
GLM 5.3 Flash is the default Agent model. It is cheap, multimodal, and spends credits on tokens and tool calls like the rest of the studio.
Best
Claude Opus 5Best
Claude Opus 5 is the Best Agent model for us: the strongest Anthropic tier we carry, for the hardest reasoning, writing, and agentic jobs.
Most cost-effective
GLM 5.3 Flash
GLM 5.3 Flash is also the cost pick: the default stays the cheapest everyday Agent model, including image and video input.
When to choose one over the other
GLM 5.3 Flash vs GPT-5.6 Luna
Use GLM 5.3 Flash for cheap multimodal Agent work. Use GPT-5.6 Luna when you want stronger writing and studio prompts at a still-low token rate.
Gemini 3.7 Flash vs DeepSeek V4 Flash
Use Gemini 3.7 Flash when the thread needs video input. Use DeepSeek V4 Flash for high-volume text-only chat when you do not need images or video.
Every current model
Prices come from the live studio catalog. The selected model stays visible while you work.
GPT-5.6 Luna
OpenAI's fast, cheap Agent model with tools, images, and files.
Everyday chat, brainstorming, and studio prompts when cost matters
- Input
- 7 credits / 1M tokens
- Output
- 40 credits / 1M tokens
Gemini 3.7 Flash
Google's fast Gemini with images, video, files, tools, and reasoning.
Quick multimodal Agent work when you need Gemini speed
- Input
- 50 credits / 1M tokens
- Output
- 250 credits / 1M tokens
Kimi K3
Moonshot's long-context model with image and video input.
Long briefs and multimodal Agent threads
- Input
- 200 credits / 1M tokens
- Output
- 1000 credits / 1M tokens
GPT-5.6 Terra
Mid-tier GPT-5.6 with tools, images, files, and extended reasoning.
Harder Agent jobs when Luna is not enough and Sol is more than you need
- Input
- 67 credits / 1M tokens
- Output
- 400 credits / 1M tokens
GPT-5.6 Sol
Flagship GPT-5.6 with the strongest reasoning in the Luna family.
Complex plans, long writing, and Agent jobs that need the top GPT tier
- Input
- 334 credits / 1M tokens
- Output
- 2000 credits / 1M tokens
Grok 4.6
xAI's Grok with tools, images, files, and extended reasoning.
Direct Agent chat with Grok's tone and tools
- Input
- 134 credits / 1M tokens
- Output
- 400 credits / 1M tokens
DeepSeek V4 Flash
Low-cost DeepSeek reasoning model for text Agent work.
High-volume text chat when multimodal input is not required
- Input
- 6 credits / 1M tokens
- Output
- 12 credits / 1M tokens
DeepSeek V4 Pro
Higher-quality DeepSeek for text Agent work.
Stronger text reasoning when Flash is not enough
- Input
- 78 credits / 1M tokens
- Output
- 156 credits / 1M tokens
MiMo V2.5
Xiaomi's multimodal model with image and video input.
Cheap multimodal Agent threads
- Input
- 10 credits / 1M tokens
- Output
- 19 credits / 1M tokens
MiniMax M3
MiniMax chat with image and video input.
Multimodal Agent work at a mid-range credit cost
- Input
- 20 credits / 1M tokens
- Output
- 80 credits / 1M tokens
Claude Sonnet 5
Anthropic's Sonnet with tools, images, files, and a 1M context window.
Writing, coding, and Agent jobs that want Claude's default quality
- Input
- 134 credits / 1M tokens
- Output
- 667 credits / 1M tokens
Claude Opus 5
Anthropic's flagship Claude with tools, images, and files.
The hardest Agent jobs when Sonnet is not enough
- Input
- 334 credits / 1M tokens
- Output
- 1667 credits / 1M tokens
GLM 5.3
Zhipu's GLM for text Agent work with a very large context window.
Long text threads when you do not need image or video input
- Input
- 94 credits / 1M tokens
- Output
- 294 credits / 1M tokens
GLM 5.3 Flash
Faster GLM with image and video input.
Cheap multimodal Agent work on the GLM family
- Input
- 5 credits / 1M tokens
- Output
- 17 credits / 1M tokens
Gemini 3.1 Pro
Google's Pro Gemini with images, video, files, and tools.
Harder multimodal Agent jobs when Flash is not enough
- Input
- 134 credits / 1M tokens
- Output
- 800 credits / 1M tokens
Qwen 3.8 Max
Alibaba's flagship Qwen with image and video input.
Multimodal Agent work when you want the top Qwen tier
- Input
- 134 credits / 1M tokens
- Output
- 400 credits / 1M tokens
Qwen 3.8 Flash
Fast Qwen with image and video input.
High-volume multimodal Agent threads at the lowest Qwen cost
- Input
- 10 credits / 1M tokens
- Output
- 32 credits / 1M tokens
Questions
- Is Agent billed like Image Studio?
- Agent spends credits on input and output tokens, plus tool calls. Image, video, and audio studios use per-generation credit rows instead.
- What does Our pick mean?
- Our pick is the default chat model for new Agent threads. You can switch models on any thread.
- Where do I chat?
- This page helps you choose. Open Agent to start a thread with the model you picked.
Open Agent
Start a thread, switch models any time, and spend credits on tokens and tool calls.