What models does Amp’s new ChatGPT-linked subscription actually includ

Watch · checked weekly

Changed on Sep 29

The old consensus stated that High dial and Oracle use GPT-6 Astra. The new consensus reverses this, stating that Low uses GPT-5.6 Terra, Medium uses GPT-5.6 Sol, and High uses GPT-5.6 Sol, while acknowledging GPT-6 Astra as a future or changing documentation name. This is a material shift in specific model assignments.

Checked 10 times since Jul 28 · paused

Direct links

A Dial for You - Amp ampcode.com

Current answer

Amp’s ChatGPT-linked subscription covers three modes:

Amp mode	Models covered by the linked ChatGPT subscription
Low	GPT-5.6 Terra
Medium	GPT-5.6 Sol
High	GPT-5.6 Sol for both the agent and oracle/reviewer
Ultra	Not included; continues using Claude Fable and Amp credits

The subscription-backed coverage also applies to supporting work such as thread reading and code review, rather than those calls silently consuming Amp credits.

One caveat is that Amp’s model routing is changing: newer documentation shows names such as GPT-6 Astra, and the exact model used—particularly for High and the Medium oracle—may vary by subscription tier and current rollout. The important practical distinction remains that Low, Medium, and High can use the linked ChatGPT allowance; Ultra and other non-eligible usage still require Amp credits.

Sources

History

How strongly the models agree, per check. Filled points mark checks that changed the answer.

Sep 29
The answer changed

The old consensus stated that High dial and Oracle use GPT-6 Astra. The new consensus reverses this, stating that Low uses GPT-5.6 Terra, Medium uses GPT-5.6 Sol, and High uses GPT-5.6 Sol, while acknowledging GPT-6 Astra as a future or changing documentation name. This is a material shift in specific model assignments.
Shown above
Sep 22
The answer changed

The routing models have fundamentally shifted. The old consensus stated that ChatGPT linking covers GPT-5.6 Terra and GPT-5.6 Sol, while Ultra and others are excluded. The new consensus updates this to include GPT-6 Astra for High-mode and Oracle roles (depending on the ChatGPT tier), replacing the older GPT-5.6 assignments for those tiers.
Read this version
Sep 1 – Sep 15 3 checks, no change
- Sep 15 Read this version
- Sep 8 Read this version
- Sep 1 Read this version
Aug 25
The answer changed

The exact model routing breakdown substantially changed: Low main agent shifted from GLM-5.2 to GPT-5.6 Terra, High oracle shifted from Claude Fable 5 to GPT-5.6 Sol, and Code review shifted from GPT-5.5 to GPT-5.6 Terra. Additionally, the new answer specifies that Low/Med/High workflows are covered, while Ultra mode uses Claude Fable and Amp credits.
Read this version
Aug 18
The answer changed

The breakdown of model assignments per routing mode changed significantly. Low mode changed from GPT-5.6 Terra to GLM-5.2 + GPT-5.6 Sol; High mode oracle changed from GPT-5.6 Sol to Claude Fable 5; Review changed from GPT-5.6 Terra to GPT-5.5; and specific routes for Puck/Search and Librarian were added while general routing details were updated.
Read this version
Aug 4 – Aug 11 2 checks, no change
- Aug 11same answer, new wording Read this version
- Aug 4 Read this version
Jul 28, 2026
Watch started

The first answer. Every later check is compared with it.
Read this version

How the 3 models compared · 1 contradiction

OpenAI GPT-5.6 Luna · Google Gemini Gemini 3.5 Flash-Lite · DeepSeek DeepSeek V4.1 Flash

contradiction · critical

High mode uses GPT-6 Astra for the agent and oracle

Model A states High mode uses GPT-6 Astra.

DeepSeek

“High** | **GPT-6 Astra** (replaces GPT-5.6 Sol) | **GPT-6 Astra”

Model B states High mode uses GPT-5.6 Sol.

OpenAI

“| GPT-5.6 Sol for both the agent and oracle”

Model C states High mode uses GPT-5.6 Sol.

Gemini

“High Mode:** Uses **GPT-5.6 Sol** for both the main agent”

How to verify: Check Amp documentation for the High mode model under a linked ChatGPT subscription.