Modes & Models

Amp uses the best model for each task: leading generalist foundation models for complex reasoning and planning, and smaller specialized models for fast, accurate responses in specific domains.

Agent Modes

Combination of System Prompt + Tools + Model

Low

Fast, low-cost mode for small, well-defined tasks

Agent GLM-5.2 · med

Oracle GPT-5.6 Sol · high

Medium

Balanced intelligence, speed, and cost for most tasks

Agent GPT-5.6 Sol · med

Oracle GPT-5.6 Sol · high

High

Deep reasoning for hard tasks

Agent GPT-5.6 Sol · x-high

Oracle Fable 5 · high

Ultra

The most capable mode for hard, open-ended tasks

Agent Fable 5 · high

Oracle GPT-5.6 Sol · high

Puck

Navigate work across Amp projects, threads, and context

GPT-5.6 Terra

Which modes are people using in Amp?

Everyone

JUL 14

JUL 15

JUL 16

JUL 17

JUL 18

JUL 19

JUL 20

JUL 21

JUL 22

JUL 23

Amp Team

JUL 14

JUL 15

JUL 16

JUL 17

JUL 18

JUL 19

JUL 20

JUL 21

JUL 22

JUL 23

low

medium

high

ultra

puck

Subagents

Review

Bug identification & code review assistance

GPT-5.5

Fast, accurate codebase retrieval

GPT-5.6 Terra

Oracle

Complex reasoning & planning on code

GPT-5.6 Sol

Librarian

Large-scale retrieval & research on external code

GPT-5.6 Sol

Read Thread

Reading & summarizing other Amp threads

GLM-5.2

System Models

View Media

Analysis of images, PDFs, audio, and video

Gemini 3 Flash

Painter

Image generation and editing

GPT Image 2

Titling

Fast title generation for threads

GPT-5.6 Luna

Compaction

Context summarization for long threads

GPT-5.6 Sol

Agent Mode Plugins

Plugins extend Amp with new tools, commands, and agent modes. These experimental modes are how we evaluate new models, and we're sharing them so you can, too. Fork and modify the plugins to run your own experiments.

Learn how plugins work

Inkling

Thinking Machines' open-weight multimodal model

Grok 4.5

xAI's coding and agent model

Claude Fable 5

Anthropic's Mythos class coding model

GLM 5.2

Z.ai's open-weight coding model, hosted by Amp

Smart

The deprecated Smart mode: Claude Opus 4.8 with high reasoning

Deep

The deprecated Deep mode: GPT-5.5 with medium reasoning

Rush

The deprecated Rush mode: GPT-5.5 with no reasoning

Large

The deprecated Large mode: long-context Claude Opus 4.8

See which models you can use in a custom agent:

Model Cards

Evals, caveats, and rollout notes behind Amp’s model choices.

GPT 5.5

It is more capable than GPT-5.4, but can cost more.

Read card

See the Amp Security Reference for more details on model use and inference providers.