Amp 2026-10-03: Opus 5.5 powers medium, faster GPT-6 Astra#

Published October 3, 2026 by Alex Sorokoletov

Part of the Amp Version Tracker series. | ampcode.com | News

Amp build 0.0.1791060199, a week on from 0.0.1790452833, ships two model changes. Claude Opus 5.5 now powers the medium mode, which Amp describes as solving more tasks for less and easier to steer, and inference for GPT-6 Astra runs six times faster.^[1]

Behind the Dial#

Amp sets model choice through the Dial, a set of notches you turn to low, medium, high, or ultra, instead of a per-request model list. You pick a tier and Amp routes the request; the model behind each tier is Amp's to change. Both of this week's updates work at that layer.

Amp ships many builds a day, so a week spans dozens of them. The two entries above are the model changes on the public changelog for this range.

What This Tells Us#

Amp keeps model choice as the Dial's notches instead of a per-request picker. Updating the model behind medium is how the change reaches people: set the Dial to medium and you now run Opus 5.5, with no model name to choose. Claude Code, Codex CLI, Gemini CLI, and OpenCode surface the model to the user through a flag, config, or a /model command. Amp's Dial keeps it behind a tier, so this swap took effect for medium users without them selecting a model.

The two updates also sit on different providers. Opus 5.5 is Anthropic's; GPT-6 Astra is OpenAI's. Amp routes across both behind the same control, so which provider serves a notch is a routing detail the user never selects. The GPT-6 Astra entry is an inference speedup rather than a model swap, so the gain there is latency: the same model, answering faster.

Sources#

1. Amp News: entries dated September 28 and September 29, 2026.